Add files using upload-large-folder tool

.gitattributes

@@ -450,3 +450,23 @@ third_person_single_vehicle_16_v1/images_1280x704/16_tp_simulation_game_maglev_p
 third_person_single_vehicle_16_v1/images_1280x704/10_tp_realistic_tundra_snowmobile.png filter=lfs diff=lfs merge=lfs -text
 third_person_single_vehicle_16_v1/images_1280x704/06_tp_realistic_valley_light_aircraft.png filter=lfs diff=lfs merge=lfs -text
 third_person_single_vehicle_16_v1/images_1280x704/09_tp_realistic_open_sea_sailboat.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/01_small_modern_studio_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/01_wide_modern_city_studio.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/08_small_fantasy_grotto_chamber.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/10_small_treehouse_platform.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/10_wide_fantasy_world_tree_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/05_small_rural_village_lane.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/04_small_desert_observatory_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/07_wide_cartoon_coastal_cafe.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/07_small_cartoon_cafe_courtyard.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/03_small_seaside_guest_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/08_wide_fantasy_floating_world_study.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/05_wide_rural_farmhouse_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/03_wide_tropical_guest_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/04_wide_desert_observatory_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/02_wide_alpine_cabin.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/09_wide_fantasy_crystal_world_shrine.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/06_small_cartoon_toy_street.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/02_small_alpine_cabin_room.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/09_small_crystal_apothecary_shop.png filter=lfs diff=lfs merge=lfs -text
+depth_scale_contrast_10x2_v1/images_1280x704/06_wide_cartoon_toy_kingdom_playroom.png filter=lfs diff=lfs merge=lfs -text

depth_scale_contrast_10x2_v1/README.md

@@ -0,0 +1,8 @@
+# Depth-Scale Contrast — 10 Wide + 10 Small
+
+Ten paired tests for physical scene-scale understanding in an action-conditioned World Model.
+
+- `wide`: aerial cities, mountain ranges, archipelagos, deserts, plains, cartoon worlds, and fantasy worlds. Large scenes are intentionally rendered with low-to-moderate spatial frequency: grouped city districts, broad vegetation masses, large terrain regions, few landmarks, atmospheric haze, and no individually resolved windows, cars, houses, trees, leaves, or repeated microstructure.
+- `small`: a deliberately mixed set of bounded small-scale environments: four rooms, two short streets, one courtyard, one pocket-like cave chamber, one compact shop, and one treehouse platform. The set avoids reducing every small space to either a room or a corridor.
+
+All images are directly generated at a 1280 × 704 final size with a bodyless observer camera. Use the identical `video_prompt_strafe` and action sequence for both images in each pair. The primary recommended action remains pure right strafe (`D`).

depth_scale_contrast_10x2_v1/depth_scale_contrast_10x2_v1.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c96da6556d20caf1f7cb14ca6d55eaf8dc9c6d7067881a823e7c73fa09ed913
+size 22532278

depth_scale_contrast_10x2_v1/manifest_summary.json

@@ -0,0 +1,20 @@
+{
+  "pairs": 10,
+  "images": 20,
+  "scale_classes": {
+    "wide": 10,
+    "small": 10
+  },
+  "styles": {
+    "realistic": 10,
+    "cartoon": 4,
+    "fantasy": 6
+  },
+  "resolution": [
+    1280,
+    704
+  ],
+  "wide_detail_standard": "low-to-moderate frequency with large grouped world regions and atmospheric simplification",
+  "small_scene_standard": "mixed bounded small-scale environments: 4 rooms, 2 streets, 1 courtyard, 1 cave chamber, 1 shop, 1 platform",
+  "recommended_primary_test": "D / pure right strafe"
+}

depth_scale_contrast_10x2_v1/metadata.csv

@@ -0,0 +1,161 @@
+id,pair_number,pair_id,scale_class,small_scene_type,style,file_name,width,height,horizontal_fov_deg,recommended_action,image_prompt,video_prompt_strafe,video_prompt_forward,expected_result
+1,1,01_modern_city_studio,wide,,realistic,images_1280x704/01_wide_modern_city_studio.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high oblique aerial view over a modern river city spanning many kilometers, with a compact skyline core, several broad residential districts, two bridges, and distant low hills under light atmospheric haze. The camera is about 200 meters above the city; all visible architecture is far below.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: soft blue-gray city blocks, muted river blue, pale warm daylight
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+2,1,01_modern_city_studio,small,room,realistic,images_1280x704/01_small_modern_studio_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a compact modern studio living room about 4 by 5 meters. From near the center, look diagonally across two adjacent close walls, a sofa, low table, compact bed, one desk, and one window. All major surfaces and furniture are one to three meters away.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: warm gray walls, medium wood, charcoal fabric, soft window daylight
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+3,2,02_alpine_cabin,wide,,realistic,images_1280x704/02_wide_alpine_cabin.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high aerial panorama over an immense alpine mountain range and deep glacial valley, with four broad ridge layers, two distant lakes, and large snow masses extending to the horizon. The nearest terrain is far below the camera, with no nearby cliff or foreground rocks.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: large white snow masses, cool blue shadows, subdued green valleys, hazy pale sky
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+4,2,02_alpine_cabin,small,room,realistic,images_1280x704/02_small_alpine_cabin_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a compact square mountain-cabin bedroom about 3.5 by 4.5 meters. The diagonal view shows two close timber walls, a low ceiling, one bed, a small table, a simple stove, and one window. Everything is contained within one small room.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: broad honey-colored timber panels, dark bed, warm lamp light, cool window light
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+5,3,03_tropical_guest_room,wide,,realistic,images_1280x704/03_wide_tropical_guest_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high aerial view over a vast tropical archipelago with six large islands, broad turquoise water channels, a distant volcanic island, and a soft horizon. Land and water form simple sweeping shapes; no aircraft, boats, nearby palm leaves, or foreground structures.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: broad turquoise water, deep blue channels, muted green island masses, pale clouds
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+6,3,03_tropical_guest_room,small,room,realistic,images_1280x704/03_small_seaside_guest_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a compact seaside guest room about 4 by 4 meters. The camera looks diagonally across two close plaster walls, a simple bed, one low cabinet, a small chair, and a broad window with soft ocean light. The complete square room is clearly visible and enclosed.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: off-white plaster, light wood, muted teal fabric, soft coastal daylight
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+7,4,04_desert_observatory_room,wide,,realistic,images_1280x704/04_wide_desert_observatory_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A vast high-altitude view across a monumental desert basin with three huge mesas, a broad dry lake, layered dunes, and mountain silhouettes many kilometers away. The land is divided into a few large smooth color regions with strong atmospheric depth and no nearby rocks or plants.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: large ochre and terracotta land masses, muted violet distance, pale sky
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+8,4,04_desert_observatory_room,small,room,realistic,images_1280x704/04_small_desert_observatory_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a small desert observatory control room about 4 meters square. The diagonal view shows two close smooth adobe walls, a compact telescope console with blank unmarked surfaces, one chair, a round ceiling opening, and a small window. All objects are large simple forms within one enclosed room.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: warm adobe, dark wood, muted brass, soft skylight
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+9,5,05_rural_farmhouse_room,wide,,realistic,images_1280x704/05_wide_rural_farmhouse_room.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high panoramic view across an enormous rural valley made of a few broad fields, a winding river, distant low mountains, and one small village cluster far below. Use large agricultural color blocks rather than individually visible crops, houses, roads, or trees.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: broad muted green and gold fields, blue river ribbon, hazy blue hills
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+10,5,05_rural_farmhouse_room,small,narrow_street,realistic,images_1280x704/05_small_rural_village_lane.png,1280,704,70,D / pure right strafe,"Use case: photorealistic-natural
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A compact old rural village lane about 3 meters wide between low stone houses, viewed from street level. Close walls and broad door shapes sit one to two meters from the camera, while the lane makes a short bend after roughly 15 meters. Use only four to six large façade masses, a simple stone road, and open sky; no people, signs, vehicles, shop clutter, or tiny masonry.
+Style/medium: soft cinematic realism with simplified large forms and restrained fine texture
+Color palette and lighting: broad cream plaster, muted gray stone, dark wood doors, warm daylight
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A short outdoor village street with close façades and a bend, not an indoor corridor or infinitely long axial alley. Keep the scene simple and locally bounded.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+11,6,06_cartoon_toy_kingdom_playroom,wide,,cartoon,images_1280x704/06_wide_cartoon_toy_kingdom_playroom.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high aerial view over a vast cheerful toy kingdom composed of five large color-block districts, a broad river, a central castle silhouette, rounded hills, and distant clouds. The entire kingdom reads at once; avoid tiny houses, windows, vehicles, characters, flags, or dense decoration.
+Style/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping
+Color palette and lighting: large pastel blue, green, coral, and yellow regions with soft cloud shapes
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+12,6,06_cartoon_toy_kingdom_playroom,small,narrow_street,cartoon,images_1280x704/06_small_cartoon_toy_street.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A compact cartoon toy-town street about 4 meters wide, enclosed by a few oversized block buildings and a rounded corner only 18 meters ahead. Close colorful façades occupy both sides, with one broad sidewalk and a clean road. No characters, cars, signs, windows grids, tiny props, or dense decorations.
+Style/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping
+Color palette and lighting: pastel blue, coral, yellow, and pale green building masses under a simple sky
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A short friendly outdoor street with close buildings and a visible corner, not a room, tunnel, hallway, or endlessly repeating corridor.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+13,7,07_cartoon_coastal_cafe,wide,,cartoon,images_1280x704/07_wide_cartoon_coastal_cafe.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high oblique animated view over a large stylized coastal town wrapped around a broad bay, with several simple neighborhood masses, two large piers, rounded hills, and open sea to the horizon. Buildings appear as clean grouped blocks, not thousands of individual structures.
+Style/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping
+Color palette and lighting: broad white and terracotta town masses, deep blue bay, muted green hills
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+14,7,07_cartoon_coastal_cafe,small,courtyard,cartoon,images_1280x704/07_small_cartoon_cafe_courtyard.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A small open-air seaside café courtyard about 6 by 6 meters, enclosed on three sides by low stucco walls with the sky visible above. The diagonal view shows a simple counter, three broad tables, a short wall fountain, and one arched opening. All walls and furniture are close; no people, signs, menus, text, or long passage.
+Style/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping
+Color palette and lighting: warm cream stucco, terracotta floor blocks, teal furniture, soft sea light
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact square courtyard with nearby enclosing walls and visible open sky; no corridor, alley, room interior, or long axial route.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+15,8,08_fantasy_floating_world_study,wide,,fantasy,images_1280x704/08_wide_fantasy_floating_world_study.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high aerial view over a colossal fantasy world of four giant floating continents, broad waterfalls, cloud oceans, one distant capital silhouette, and mountain ranges extending to the horizon. Use very large clean landforms and only a few landmarks; no nearby platform, creature, railing, or small floating debris.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: large emerald land masses, blue-white clouds, pale stone landmarks, soft golden light
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+16,8,08_fantasy_floating_world_study,small,cave_chamber,fantasy,images_1280x704/08_small_fantasy_grotto_chamber.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a compact natural fantasy grotto about 5 meters across. Close rounded rock walls enclose a shallow pool, two large luminous plants, three broad stepping stones, and one small skylight opening above. The entire pocket-like cave chamber is visible at once; no tunnel mouth, branching passage, corridor, tiny crystals, or particles.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: smooth dark rock, muted emerald pool, two soft cyan plant lights, warm skylight
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A small rounded cave chamber or pocket, not a tunnel, corridor, passage, mine shaft, or long cavern route.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+17,9,09_fantasy_crystal_world_shrine,wide,,fantasy,images_1280x704/09_wide_fantasy_crystal_world_shrine.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A vast aerial panorama across a crystal world with three enormous smooth crystal ridges, a broad luminous lake, two dark plateaus, and a glowing horizon. Crystal formations are giant and widely separated; do not fill the image with tiny shards, sparkles, particles, or intricate facets.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: large deep-blue land masses, cyan crystal planes, muted violet distance
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+18,9,09_fantasy_crystal_world_shrine,small,shop,fantasy,images_1280x704/09_small_crystal_apothecary_shop.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: Inside a compact fantasy crystal apothecary shop about 4 by 4.5 meters. A diagonal view shows two close walls, one broad counter, two simplified shelves with large bottle silhouettes, three large crystal lamps, and a round window. No people, labels, readable text, dense bottles, long aisle, or doorway-centered view.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: dark smooth wood and stone, three cyan crystal masses, restrained violet glow
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact single shop space viewed diagonally, with a close counter and walls; no corridor, long aisle, warehouse, or doorway tunnel.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close
+19,10,10_fantasy_world_tree_room,wide,,fantasy,images_1280x704/10_wide_fantasy_world_tree_room.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A high aerial panorama above the canopy of a colossal world tree whose crown spans an entire valley, with five giant branch masses, distant mountains, cloud layers, and a few large suspended gardens. Treat foliage as broad continuous canopy shapes, not individually rendered leaves or countless branches.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: broad deep-green canopy masses, warm wood branches, blue haze, pale clouds
+Camera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.
+Scale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",smaller optical-flow magnitude because dominant visible geometry is far away
+20,10,10_fantasy_world_tree_room,small,platform,fantasy,images_1280x704/10_small_treehouse_platform.png,1280,704,70,D / pure right strafe,"Use case: stylized-concept
+Asset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment
+Primary request: A compact circular treehouse platform about 5 meters across wrapped around one enormous tree trunk. The trunk fills one side at close range, a waist-high wooden railing encloses the platform, a small canopy covers half the deck, and one built-in bench and table sit nearby. The platform is locally bounded despite the distant forest view; no bridge, corridor, ladder shaft, or long walkway.
+Style/medium: cinematic fantasy environment art with clean major forms and restrained magical accents
+Color palette and lighting: warm broad wood surfaces, muted moss-green canopy, soft golden forest light
+Camera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.
+Scale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact bounded platform with close trunk, railing, furniture, and canopy; no bridge, corridor, walkway, room interior, or long route.
+Global constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.","The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.","The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.",larger optical-flow magnitude because room surfaces and furniture are close

depth_scale_contrast_10x2_v1/metadata.jsonl

@@ -0,0 +1,20 @@
+{"id": 1, "pair_number": 1, "pair_id": "01_modern_city_studio", "scale_class": "wide", "small_scene_type": "", "style": "realistic", "file_name": "images_1280x704/01_wide_modern_city_studio.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high oblique aerial view over a modern river city spanning many kilometers, with a compact skyline core, several broad residential districts, two bridges, and distant low hills under light atmospheric haze. The camera is about 200 meters above the city; all visible architecture is far below.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: soft blue-gray city blocks, muted river blue, pale warm daylight\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 2, "pair_number": 1, "pair_id": "01_modern_city_studio", "scale_class": "small", "small_scene_type": "room", "style": "realistic", "file_name": "images_1280x704/01_small_modern_studio_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a compact modern studio living room about 4 by 5 meters. From near the center, look diagonally across two adjacent close walls, a sofa, low table, compact bed, one desk, and one window. All major surfaces and furniture are one to three meters away.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: warm gray walls, medium wood, charcoal fabric, soft window daylight\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 3, "pair_number": 2, "pair_id": "02_alpine_cabin", "scale_class": "wide", "small_scene_type": "", "style": "realistic", "file_name": "images_1280x704/02_wide_alpine_cabin.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high aerial panorama over an immense alpine mountain range and deep glacial valley, with four broad ridge layers, two distant lakes, and large snow masses extending to the horizon. The nearest terrain is far below the camera, with no nearby cliff or foreground rocks.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: large white snow masses, cool blue shadows, subdued green valleys, hazy pale sky\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 4, "pair_number": 2, "pair_id": "02_alpine_cabin", "scale_class": "small", "small_scene_type": "room", "style": "realistic", "file_name": "images_1280x704/02_small_alpine_cabin_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a compact square mountain-cabin bedroom about 3.5 by 4.5 meters. The diagonal view shows two close timber walls, a low ceiling, one bed, a small table, a simple stove, and one window. Everything is contained within one small room.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: broad honey-colored timber panels, dark bed, warm lamp light, cool window light\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 5, "pair_number": 3, "pair_id": "03_tropical_guest_room", "scale_class": "wide", "small_scene_type": "", "style": "realistic", "file_name": "images_1280x704/03_wide_tropical_guest_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high aerial view over a vast tropical archipelago with six large islands, broad turquoise water channels, a distant volcanic island, and a soft horizon. Land and water form simple sweeping shapes; no aircraft, boats, nearby palm leaves, or foreground structures.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: broad turquoise water, deep blue channels, muted green island masses, pale clouds\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 6, "pair_number": 3, "pair_id": "03_tropical_guest_room", "scale_class": "small", "small_scene_type": "room", "style": "realistic", "file_name": "images_1280x704/03_small_seaside_guest_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a compact seaside guest room about 4 by 4 meters. The camera looks diagonally across two close plaster walls, a simple bed, one low cabinet, a small chair, and a broad window with soft ocean light. The complete square room is clearly visible and enclosed.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: off-white plaster, light wood, muted teal fabric, soft coastal daylight\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 7, "pair_number": 4, "pair_id": "04_desert_observatory_room", "scale_class": "wide", "small_scene_type": "", "style": "realistic", "file_name": "images_1280x704/04_wide_desert_observatory_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A vast high-altitude view across a monumental desert basin with three huge mesas, a broad dry lake, layered dunes, and mountain silhouettes many kilometers away. The land is divided into a few large smooth color regions with strong atmospheric depth and no nearby rocks or plants.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: large ochre and terracotta land masses, muted violet distance, pale sky\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 8, "pair_number": 4, "pair_id": "04_desert_observatory_room", "scale_class": "small", "small_scene_type": "room", "style": "realistic", "file_name": "images_1280x704/04_small_desert_observatory_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a small desert observatory control room about 4 meters square. The diagonal view shows two close smooth adobe walls, a compact telescope console with blank unmarked surfaces, one chair, a round ceiling opening, and a small window. All objects are large simple forms within one enclosed room.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: warm adobe, dark wood, muted brass, soft skylight\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A contained room viewed diagonally; no corridor, hallway, tunnel, passage, alley, aisle, or doorway-centered long axis.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 9, "pair_number": 5, "pair_id": "05_rural_farmhouse_room", "scale_class": "wide", "small_scene_type": "", "style": "realistic", "file_name": "images_1280x704/05_wide_rural_farmhouse_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high panoramic view across an enormous rural valley made of a few broad fields, a winding river, distant low mountains, and one small village cluster far below. Use large agricultural color blocks rather than individually visible crops, houses, roads, or trees.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: broad muted green and gold fields, blue river ribbon, hazy blue hills\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 10, "pair_number": 5, "pair_id": "05_rural_farmhouse_room", "scale_class": "small", "small_scene_type": "narrow_street", "style": "realistic", "file_name": "images_1280x704/05_small_rural_village_lane.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: photorealistic-natural\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A compact old rural village lane about 3 meters wide between low stone houses, viewed from street level. Close walls and broad door shapes sit one to two meters from the camera, while the lane makes a short bend after roughly 15 meters. Use only four to six large façade masses, a simple stone road, and open sky; no people, signs, vehicles, shop clutter, or tiny masonry.\nStyle/medium: soft cinematic realism with simplified large forms and restrained fine texture\nColor palette and lighting: broad cream plaster, muted gray stone, dark wood doors, warm daylight\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A short outdoor village street with close façades and a bend, not an indoor corridor or infinitely long axial alley. Keep the scene simple and locally bounded.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 11, "pair_number": 6, "pair_id": "06_cartoon_toy_kingdom_playroom", "scale_class": "wide", "small_scene_type": "", "style": "cartoon", "file_name": "images_1280x704/06_wide_cartoon_toy_kingdom_playroom.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high aerial view over a vast cheerful toy kingdom composed of five large color-block districts, a broad river, a central castle silhouette, rounded hills, and distant clouds. The entire kingdom reads at once; avoid tiny houses, windows, vehicles, characters, flags, or dense decoration.\nStyle/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping\nColor palette and lighting: large pastel blue, green, coral, and yellow regions with soft cloud shapes\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 12, "pair_number": 6, "pair_id": "06_cartoon_toy_kingdom_playroom", "scale_class": "small", "small_scene_type": "narrow_street", "style": "cartoon", "file_name": "images_1280x704/06_small_cartoon_toy_street.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A compact cartoon toy-town street about 4 meters wide, enclosed by a few oversized block buildings and a rounded corner only 18 meters ahead. Close colorful façades occupy both sides, with one broad sidewalk and a clean road. No characters, cars, signs, windows grids, tiny props, or dense decorations.\nStyle/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping\nColor palette and lighting: pastel blue, coral, yellow, and pale green building masses under a simple sky\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A short friendly outdoor street with close buildings and a visible corner, not a room, tunnel, hallway, or endlessly repeating corridor.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 13, "pair_number": 7, "pair_id": "07_cartoon_coastal_cafe", "scale_class": "wide", "small_scene_type": "", "style": "cartoon", "file_name": "images_1280x704/07_wide_cartoon_coastal_cafe.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high oblique animated view over a large stylized coastal town wrapped around a broad bay, with several simple neighborhood masses, two large piers, rounded hills, and open sea to the horizon. Buildings appear as clean grouped blocks, not thousands of individual structures.\nStyle/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping\nColor palette and lighting: broad white and terracotta town masses, deep blue bay, muted green hills\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 14, "pair_number": 7, "pair_id": "07_cartoon_coastal_cafe", "scale_class": "small", "small_scene_type": "courtyard", "style": "cartoon", "file_name": "images_1280x704/07_small_cartoon_cafe_courtyard.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A small open-air seaside café courtyard about 6 by 6 meters, enclosed on three sides by low stucco walls with the sky visible above. The diagonal view shows a simple counter, three broad tables, a short wall fountain, and one arched opening. All walls and furniture are close; no people, signs, menus, text, or long passage.\nStyle/medium: polished stylized 3D animation environment with clean silhouettes and controlled color grouping\nColor palette and lighting: warm cream stucco, terracotta floor blocks, teal furniture, soft sea light\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact square courtyard with nearby enclosing walls and visible open sky; no corridor, alley, room interior, or long axial route.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 15, "pair_number": 8, "pair_id": "08_fantasy_floating_world_study", "scale_class": "wide", "small_scene_type": "", "style": "fantasy", "file_name": "images_1280x704/08_wide_fantasy_floating_world_study.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high aerial view over a colossal fantasy world of four giant floating continents, broad waterfalls, cloud oceans, one distant capital silhouette, and mountain ranges extending to the horizon. Use very large clean landforms and only a few landmarks; no nearby platform, creature, railing, or small floating debris.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: large emerald land masses, blue-white clouds, pale stone landmarks, soft golden light\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 16, "pair_number": 8, "pair_id": "08_fantasy_floating_world_study", "scale_class": "small", "small_scene_type": "cave_chamber", "style": "fantasy", "file_name": "images_1280x704/08_small_fantasy_grotto_chamber.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a compact natural fantasy grotto about 5 meters across. Close rounded rock walls enclose a shallow pool, two large luminous plants, three broad stepping stones, and one small skylight opening above. The entire pocket-like cave chamber is visible at once; no tunnel mouth, branching passage, corridor, tiny crystals, or particles.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: smooth dark rock, muted emerald pool, two soft cyan plant lights, warm skylight\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A small rounded cave chamber or pocket, not a tunnel, corridor, passage, mine shaft, or long cavern route.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 17, "pair_number": 9, "pair_id": "09_fantasy_crystal_world_shrine", "scale_class": "wide", "small_scene_type": "", "style": "fantasy", "file_name": "images_1280x704/09_wide_fantasy_crystal_world_shrine.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A vast aerial panorama across a crystal world with three enormous smooth crystal ridges, a broad luminous lake, two dark plateaus, and a glowing horizon. Crystal formations are giant and widely separated; do not fill the image with tiny shards, sparkles, particles, or intricate facets.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: large deep-blue land masses, cyan crystal planes, muted violet distance\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 18, "pair_number": 9, "pair_id": "09_fantasy_crystal_world_shrine", "scale_class": "small", "small_scene_type": "shop", "style": "fantasy", "file_name": "images_1280x704/09_small_crystal_apothecary_shop.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: Inside a compact fantasy crystal apothecary shop about 4 by 4.5 meters. A diagonal view shows two close walls, one broad counter, two simplified shelves with large bottle silhouettes, three large crystal lamps, and a round window. No people, labels, readable text, dense bottles, long aisle, or doorway-centered view.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: dark smooth wood and stone, three cyan crystal masses, restrained violet glow\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact single shop space viewed diagonally, with a close counter and walls; no corridor, long aisle, warehouse, or doorway tunnel.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}
+{"id": 19, "pair_number": 10, "pair_id": "10_fantasy_world_tree_room", "scale_class": "wide", "small_scene_type": "", "style": "fantasy", "file_name": "images_1280x704/10_wide_fantasy_world_tree_room.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A high aerial panorama above the canopy of a colossal world tree whose crown spans an entire valley, with five giant branch masses, distant mountains, cloud layers, and a few large suspended gardens. Treat foliage as broad continuous canopy shapes, not individually rendered leaves or countless branches.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: broad deep-green canopy masses, warm wood branches, blue haze, pale clouds\nCamera and composition: High oblique aerial viewpoint with all dominant terrain or architecture far below or far away; stable broad horizon.\nScale and detail constraints: Wide-scene frequency constraint: aggressively suppress high-frequency detail even though the world is huge. Render cities as a few continuous district masses, vegetation as broad canopy or field color blocks, and geology as large smooth landforms. Do not resolve individual windows, cars, people, street furniture, roof tiles, houses, trees, leaves, crop rows, gravel, tiny rocks, or dense repeated structures. Limit the composition to roughly five to twelve large landmarks or regions. Use gentle atmospheric haze and broad tonal gradients to soften distant edges. The result must remain visually substantial and clearly large-scale, but its dominant information should be low and mid spatial frequency rather than microtexture.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "smaller optical-flow magnitude because dominant visible geometry is far away"}
+{"id": 20, "pair_number": 10, "pair_id": "10_fantasy_world_tree_room", "scale_class": "small", "small_scene_type": "platform", "style": "fantasy", "file_name": "images_1280x704/10_small_treehouse_platform.png", "width": 1280, "height": 704, "horizontal_fov_deg": 70, "recommended_action": "D / pure right strafe", "image_prompt": "Use case: stylized-concept\nAsset type: controlled first frame for a physical scene-scale and optical-flow World Model experiment\nPrimary request: A compact circular treehouse platform about 5 meters across wrapped around one enormous tree trunk. The trunk fills one side at close range, a waist-high wooden railing encloses the platform, a small canopy covers half the deck, and one built-in bench and table sit nearby. The platform is locally bounded despite the distant forest view; no bridge, corridor, ladder shaft, or long walkway.\nStyle/medium: cinematic fantasy environment art with clean major forms and restrained magical accents\nColor palette and lighting: warm broad wood surfaces, muted moss-green canopy, soft golden forest light\nCamera and composition: Bodyless eye-level or locally appropriate viewpoint inside the bounded small scene; choose an oblique composition that clearly shows nearby surfaces and the finite local layout.\nScale and detail constraints: Small-scene frequency constraint: make the local physical scale unmistakably small using close surfaces, nearby large objects, and a clearly bounded space. Use four to eight large simple environmental elements, broad material regions, sparse decoration, and low-to-moderate detail. Do not add dense clutter, high-frequency textures, tiny repeated objects, crowds, or text. A compact bounded platform with close trunk, railing, furniture, and canopy; no bridge, corridor, walkway, room interior, or long route.\nGlobal constraints: Bodyless first-person observer; no hands, arms, feet, body, vehicle interior, drone body, camera rig, operator shadow, or reflection. Rectilinear 70-degree horizontal FOV, no fisheye, sharp stable frame, no motion blur, no shallow depth of field. No people, readable text, letters, numbers, logos, watermark, border, collage, split screen, or UI. Generate directly from scratch for a 1280x704-safe wide crop.", "video_prompt_strafe": "The camera translates exactly one meter to the right at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "video_prompt_forward": "The camera translates exactly one meter straight forward at constant speed over two seconds. Keep yaw, pitch, roll, focal length, and camera altitude fixed. Every scene object remains static. No cuts, zoom, shake, motion blur, new objects, or geometry changes.", "expected_result": "larger optical-flow magnitude because room surfaces and furniture are close"}