Promote Raven quality stack

README.md

@@ -10,10 +10,16 @@ pinned: false
 license: apache-2.0
 short_description: Governed gothic couture visual creation command center
 models:
+  - black-forest-labs/FLUX.2-klein-9B
   - black-forest-labs/FLUX.2-klein-4B
+  - Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf
   - nvidia/LocateAnything-3B
   - openbmb/MiniCPM-V-4.6
   - nvidia/NVIDIA-Nemotron-Parse-v1.2
+  - openbmb/MiniCPM5-1B
+  - onnx-community/functiongemma-270m-it-ONNX
+  - hexgrad/Kokoro-82M
+  - netflix/void-model
 tags:
   - gradio
   - mcp-server
@@ -24,6 +30,8 @@ tags:
   - best-agent
   - best-demo
   - openbmb
+  - nvidia
+  - modal
   - codex
 ---
 
@@ -49,14 +57,17 @@ The interface is built around a command-center surface:
 
 Pinned lanes do not rotate:
 
-- `image_generation`: public-demo FLUX.2 Klein 4B image lane
+- `image_generation`: Raven Quality Stack with `black-forest-labs/FLUX.2-klein-9B` as the flagship image/edit lane
 - `grounding`: NVIDIA LocateAnything-3B grounding anchor
 - `security`: ST3GG defensive scanner/export gate
 
 Sponsor/evidence lanes are optional but first-class when secrets are configured:
 
+- `Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf` (11.91B): quality/taste/lore critique lane for private or configured judge use.
 - `openbmb/MiniCPM-V-4.6` (1.30B): visual judge for wardrobe, footwear, material drift, lore continuity, and export notes.
 - `nvidia/NVIDIA-Nemotron-Parse-v1.2` (0.94B): structured evidence/parser lane for NVIDIA/Nemotron claim support.
+- `hexgrad/Kokoro-82M` (0.082B): optional lore narration lane.
+- `netflix/void-model` (5B CogVideoX-based): Modal/offline video repair sample lane, not a blocking Space runtime default.
 
 Helper lanes may rotate with quota, license, health, and parameter-budget checks:
 
@@ -67,12 +78,12 @@ Helper lanes may rotate with quota, license, health, and parameter-budget checks
 - HF catalog research
 - Modal job runner
 
-Public demo mode excludes private, commercial-uncleared, and research-only helper models. Private research mode keeps the gated FLUX.2 Klein 9B and OFFELLIA/Gemma routes available, but it never disables consent, provenance, ST3GG, export, or dataset-partition gates.
+The default preset is **Raven Quality Stack**. `black-forest-labs/FLUX.2-klein-4B` remains available only as a Tiny Titan/public-safe sidecar and fallback if the gated 9B lane is unavailable. OFFELLIA heretic/obliterated-style variants stay private research only and never disable consent, provenance, ST3GG, export, or dataset-partition gates.
 
 ## Current Features
 
 - Gradio Blocks dashboard with split update regions.
-- Real FLUX.2 Klein 4B image generation on Hugging Face ZeroGPU when runtime access is configured.
+- Real FLUX.2 Klein 9B-first image generation on Hugging Face ZeroGPU when runtime access is configured, with an honest 4B sidecar fallback.
 - Above-fold ST3GG trust strip with safe-vs-blocked fixture evidence.
 - Generated artifact ST3GG scan and checkpoint/export state.
 - Optional MiniCPM-V and Nemotron provider evidence lanes with explicit configured/missing-secret status.
@@ -89,15 +100,16 @@ Public demo mode excludes private, commercial-uncleared, and research-only helpe
 | Target | Evidence status |
 | --- | --- |
 | Gradio Space | App runs as a public Hugging Face Gradio Space with `mcp_server=True`. |
-| <=32B models | Public stack is 11.42B: FLUX.2 Klein 4B + LocateAnything 3.83B + MiniCPM-V 1.30B + Nemotron Parse 0.94B + MiniCPM5 1.08B + FunctionGemma 0.27B. |
+| <=32B models | Raven Quality Stack is 28.50B: FLUX.2 Klein 9B + OFFELLIA Q4 Gemma 12B + LocateAnything 3.83B + MiniCPM-V 1.30B + Nemotron Parse 0.94B + MiniCPM5 1.08B + FunctionGemma 0.27B + Kokoro 0.082B. |
 | Off Brand | Custom command-center UI, dense inspector, workflow graph, wardrobe/lore drawer, and provider cards. |
 | Best Agent | Multi-step prompt, generation, scan, judge, checkpoint, export workflow. |
 | OpenBMB | Claimed only when MiniCPM-V returns success status in an export packet. |
 | NVIDIA | Claimed only when Nemotron returns success status in an export packet. LocateAnything remains visible but is not the Nemotron claim by itself. |
+| Modal | Sidecar-only until a real Modal job is documented; target lane is `netflix/void-model` video repair. |
 | OpenAI Codex | Development branch and PR include Codex-authored implementation commits. |
 | Demo / social | Add final links here before submission: `DEMO_VIDEO_URL` and `SOCIAL_POST_URL`. |
 
-Tiny Titan can be claimed only from a successful public-demo export packet because the active public models are each <=4B. The stronger FLUX.2 Klein 9B and OFFELLIA/Gemma stack remains private research only.
+Tiny Titan is not the flagship story. It can be claimed only from a successful sidecar export packet where every active sidecar model is <=4B.
 
 ## Local Setup
 

app.py

@@ -49,10 +49,32 @@ MODEL_RELAY = WeaverModelRelay()
 
 def _default_operator_state() -> dict[str, Any]:
     return {
+        "active_preset": "Raven Quality Stack",
         "provider_state": "idle",
         "checkpoint": "pending",
         "export": "pending",
         "message": "No operator action yet.",
+        "modal_video_repair": {
+            "status": "deferred",
+            "repo_id": "netflix/void-model",
+            "provider": "modal",
+            "message": "Offline/Modal VOID repair sample is documented but not a blocking runtime default.",
+        },
+        "offellia_judge": {
+            "status": "deferred_local",
+            "repo_id": "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
+            "message": "Quality/taste judge lane is declared in the Raven stack and runs only when local GGUF runtime is configured.",
+        },
+        "audio_lore_tts": {
+            "status": "optional",
+            "repo_id": "hexgrad/Kokoro-82M",
+            "message": "Lore narration lane is off by default for the public demo.",
+        },
+        "tiny_titan_sidecar": {
+            "status": "available",
+            "repo_id": "black-forest-labs/FLUX.2-klein-4B",
+            "message": "Public-safe 4B sidecar remains selectable without weakening the quality preset.",
+        },
     }
 
 
@@ -163,6 +185,8 @@ def run_weave(
     if generation.status == "success":
         provider_state = "generated"
     operator_state = {
+        **_default_operator_state(),
+        "active_preset": "Raven Quality Stack",
         "provider_state": provider_state,
         "checkpoint": "pending_review",
         "export": generated_scan.get("export_gate", "pending"),
@@ -172,6 +196,20 @@ def run_weave(
         "generated_scan": generated_scan,
         "minicpm_judge": minicpm.to_dict(),
         "nemotron_evidence": nemotron.to_dict(),
+        "locateanything_grounding": {
+            "status": run.inspection.status,
+            "repo_id": run.inspection.locate_model,
+            "targets": [
+                {
+                    "slot_name": target.slot_name,
+                    "query": target.query,
+                    "expected_region": target.expected_region,
+                    "confidence": target.confidence,
+                }
+                for target in run.inspection.targets[:6]
+            ],
+            "drift_flags": run.inspection.drift_flags,
+        },
     }
     regions = _dashboard_regions(
         run=run,
@@ -304,9 +342,27 @@ def scan_reference(
             scan=reference_scan,
             wardrobe_summary=_wardrobe_summary(run),
         )
+    locate_plan = {}
+    if run is not None:
+        locate_plan = {
+            "status": run.inspection.status,
+            "repo_id": run.inspection.locate_model,
+            "source": "reference_scan",
+            "targets": [
+                {
+                    "slot_name": target.slot_name,
+                    "query": target.query,
+                    "expected_region": target.expected_region,
+                    "confidence": target.confidence,
+                }
+                for target in run.inspection.targets[:6]
+            ],
+            "drift_flags": run.inspection.drift_flags,
+        }
     next_state = {
         **state,
         **({"reference_judge": minicpm.to_dict()} if minicpm else {}),
+        **({"reference_locate_plan": locate_plan, "locateanything_grounding": locate_plan} if locate_plan else {}),
         "reference_scan": reference_scan,
         "reference_export_gate": reference_scan.get("export_gate", "pending"),
         "export": state.get("export", generated_scan.get("export_gate", "pending")),

docs/HACKATHON_EVALUATION.md

@@ -13,11 +13,11 @@ NEXUS Visual Weaver should open as a working command center, not a landing page.
 ## Current Strengths
 
 - Gradio-compatible app shape with `mcp_server=True`.
-- Pinned model governance is visible: FLUX.2 Klein 4B, LocateAnything-3B, and ST3GG.
-- Real FLUX.2 Klein 4B generation is wired for HF Space and falls back to an honest dry-run state outside Space.
+- Pinned model governance is visible: FLUX.2 Klein 9B, LocateAnything-3B, and ST3GG.
+- Real FLUX.2 Klein 9B-first generation is wired for HF Space and falls back to an honest 4B Tiny Titan sidecar when the gated lane is unavailable.
 - Generated artifacts are scanned by ST3GG before checkpoint/export.
 - Above-fold trust strip makes ST3GG verdict, export gate, fixture evidence, and adult-mode safety boundaries visible immediately.
-- OpenBMB MiniCPM-V 4.6 and NVIDIA Nemotron evidence lanes are represented as real optional provider adapters with missing-secret/failed/success states.
+- OpenBMB MiniCPM-V 4.6, NVIDIA Nemotron, OFFELLIA Q4, LocateAnything, Kokoro TTS, and Modal VOID evidence lanes are represented with missing-secret/deferred/failed/success states.
 - Adult Mode starts off and is framed as catalog scope, not a safety bypass.
 - ModelRelay/GMR helper rotation is represented without replacing pinned lanes.
 - Tests cover catalog scope, workflow planning, ModelRelay behavior, scanner evidence, and dashboard fallback rendering.
@@ -32,21 +32,21 @@ NEXUS Visual Weaver should open as a working command center, not a landing page.
 
 ## Next Implementation Priority
 
-1. Configure OpenBMB and Nemotron Space secrets if prize claims are desired.
-2. Run one live Space weave and prepare an export packet.
-3. Capture demo video and create social post.
-4. Add final demo/social URLs to README.
-5. Add Playwright/browser visual checks for desktop and mobile overflow once CI is unblocked.
+1. Keep Raven Quality Stack as the submission narrative; use Tiny Titan only as a sidecar export.
+2. Configure OpenBMB and Nemotron Space secrets if sponsor prize claims are desired.
+3. Run one live Space weave and prepare an export packet.
+4. Run/document one Modal sidecar job only if it can complete without risking the main Space.
+5. Capture demo video, create social post, and add final links to README.
 
 ## Prize Claim Evidence Rules
 
 | Prize or badge | Current stance |
 | --- | --- |
-| Build Small base eligibility | Gradio Space, <=32B stack, and public app path are ready; demo/social links still required. |
+| Build Small base eligibility | Gradio Space, each active model <32B, and public app path are ready; demo/social links still required. |
 | Off Brand | Strong custom command-center UI signal. |
 | Best Agent | Multi-step governed workflow is implemented through callbacks and export packet. |
 | OpenBMB | Claim only after MiniCPM-V returns `success` in export evidence. |
 | NVIDIA | Claim only after Nemotron returns `success` in export evidence. |
 | OpenAI Codex | GitHub branch/PR provides Codex development trail. |
-| Tiny Titan | Public demo stack is eligible: active public models are each <=4B. |
-| Modal | Not claimed unless a real Modal job runs. |
+| Tiny Titan | Sidecar-only: claim only from an export packet where every active sidecar model is <=4B. |
+| Modal | Not claimed unless a real `netflix/void-model` or equivalent Modal job runs and is documented. |

docs/HANDOFF_FINAL_HACKATHON.md

@@ -8,19 +8,20 @@
 - Public Space URL: `https://build-small-hackathon-nexus-visual-weaver-a107340.hf.space/`
 - HF rollback SHA: `410a467c55d11e7308249198bd5fe0b2c190aec6`.
 - Branch discipline: use only `main` and `codex/specimba/ui-polish-command-center`; no extra recovery branches.
-- Primary goal: finish a countable Build Small submission with real FLUX.2 4B generation, ST3GG scan, optional OpenBMB MiniCPM-V judge evidence, optional NVIDIA Nemotron evidence, checkpointed export packet, README prize mapping, demo video, and social post.
+- Primary goal: finish a countable Build Small submission with Raven Quality Stack generation, ST3GG scan, LocateAnything grounding, optional OpenBMB MiniCPM-V judge evidence, optional NVIDIA Nemotron evidence, optional Modal VOID sidecar evidence, checkpointed export packet, README prize mapping, demo video, and social post.
 
 ## Secrets Needed
 
 Do not paste these into chat, commits, logs, or export packets.
 
-- `HF_TOKEN`: optional for public FLUX.2 Klein 4B access and required only if private/gated research lanes are enabled.
+- `HF_TOKEN`: required for gated FLUX.2 Klein 9B access after license acceptance; the app can honestly fall back to the 4B Tiny Titan sidecar if the 9B lane is unavailable.
 - `MINICPM_BASE_URL`: OpenBMB OpenAI-compatible endpoint base URL.
 - `MINICPM_API_KEY`: OpenBMB bearer token.
 - `MINICPM_MODEL`: default `MiniCPM-V-4.6`.
 - `NEMOTRON_BASE_URL`: OpenAI-compatible Nemotron endpoint if available.
 - `NEMOTRON_API_KEY` or `NVIDIA_API_KEY`: Nemotron provider token.
 - `NEMOTRON_MODEL`: default `nvidia/NVIDIA-Nemotron-Parse-v1.2`.
+- `MODAL_TOKEN_ID` and `MODAL_TOKEN_SECRET`: optional for a documented `netflix/void-model` video repair sidecar job.
 
 ## Verification Commands
 
@@ -59,27 +60,28 @@ Current evidence from the SSE API:
 
 ## Runtime Flow
 
-1. `run_active_weave` builds the Raven Chronicle run packet.
-2. FLUX.2 Klein 4B generates the image on Space when HF runtime is enabled.
+1. `run_active_weave` builds the Raven Chronicle run packet from prompt, wardrobe, lore, model stack, and LocateAnything region plan.
+2. FLUX.2 Klein 9B generates the flagship image on Space when HF runtime and gated access are configured; FLUX.2 Klein 4B is an honest sidecar fallback.
 3. Generated artifact is scanned by ST3GG.
 4. MiniCPM-V judge runs when OpenBMB secrets are present.
 5. Nemotron evidence runs when Nemotron/NVIDIA endpoint secrets are present.
-6. `approve_checkpoint` requires a generated artifact and ST3GG clear/pass state.
-7. `prepare_export_packet` writes a governed JSON packet to `/data/nexus_visual_weaver/exports` or `outputs/exports`.
+6. Modal VOID repair remains a sidecar evidence lane until a real job is documented.
+7. `approve_checkpoint` requires a generated artifact and ST3GG clear/pass state.
+8. `prepare_export_packet` writes a governed JSON packet to `/data/nexus_visual_weaver/exports` or `outputs/exports`.
 
 ## Claim Rules
 
 - OpenBMB prize claim requires `minicpm_judge.status == "success"` in an export packet.
 - NVIDIA prize claim requires `nemotron_evidence.status == "success"` in an export packet.
 - LocateAnything supports the grounding story but does not replace Nemotron for the NVIDIA prize.
-- Tiny Titan can be claimed only from a successful public-demo export packet because each active public model is <=4B.
-- FLUX.2 Klein 9B and OFFELLIA/Gemma remain private research options only.
-- Modal is not claimed unless a real Modal job runs and is documented.
+- Tiny Titan can be claimed only from a successful sidecar export packet because each active sidecar model is <=4B.
+- Raven Quality Stack is the primary story: FLUX.2 Klein 9B, OFFELLIA Q4, LocateAnything, MiniCPM-V, Nemotron, MiniCPM5, FunctionGemma, and Kokoro are individually under 32B.
+- Modal is not claimed unless a real `netflix/void-model` or equivalent Modal job runs and is documented.
 
 ## Known Risks
 
 - GitHub CLI may fail behind proxy `127.0.0.1:9`; use local git status and HF verification when blocked.
-- Real FLUX generation depends on Space GPU availability and the public 4B runtime loading successfully.
+- Real FLUX generation depends on Space GPU availability and gated 9B access; the 4B sidecar exists to keep the demo useful without mislabeling the flagship lane.
 - OpenBMB and Nemotron endpoints are optional and must show `missing secret` rather than fake success when not configured.
 - Demo video and social post links must be added before final submission.
 

src/nexus_visual_weaver/catalog.py

@@ -7,32 +7,30 @@ from .schema import AdapterRecipe, ModelCandidate
 MODEL_CATALOG: list[ModelCandidate] = [
     ModelCandidate(
         repo_id="black-forest-labs/FLUX.2-klein-4B",
-        role="image_generator",
+        role="tiny_titan_sidecar_image_generator",
         task="image-to-image",
         params_b=4.0,
-        runtime="diffusers / provider",
+        runtime="diffusers / public fallback",
         license="apache-2.0",
         source_url="https://hf.co/black-forest-labs/FLUX.2-klein-4B",
     ),
     ModelCandidate(
         repo_id="black-forest-labs/FLUX.2-klein-9B",
-        role="private_research_image_generator",
+        role="quality_image_generator",
         task="image-to-image",
         params_b=9.0,
-        runtime="diffusers / gated provider",
+        runtime="diffusers / gated quality lane",
         license="other",
         gated=True,
-        public_demo=False,
         source_url="https://hf.co/black-forest-labs/FLUX.2-klein-9B",
     ),
     ModelCandidate(
         repo_id="Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
-        role="private_research_multimodal_judge",
+        role="quality_multimodal_judge",
         task="image-text-to-text",
         params_b=12.0,
         runtime="llama.cpp GGUF",
         license="apache-2.0",
-        public_demo=False,
         source_url="https://hf.co/Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
     ),
     ModelCandidate(
@@ -82,7 +80,7 @@ MODEL_CATALOG: list[ModelCandidate] = [
     ),
     ModelCandidate(
         repo_id="Brunobkr/OFFELLIA_IQ4_XS_gemma-4-12B-it-heretic",
-        role="adult_mode_text_judge",
+        role="adult_private_research_text_judge",
         task="text-generation",
         params_b=12.0,
         runtime="llama.cpp GGUF",
@@ -90,6 +88,34 @@ MODEL_CATALOG: list[ModelCandidate] = [
         adult_only=True,
         source_url="https://hf.co/Brunobkr/OFFELLIA_IQ4_XS_gemma-4-12B-it-heretic",
     ),
+    ModelCandidate(
+        repo_id="hexgrad/Kokoro-82M",
+        role="audio_lore_tts",
+        task="text-to-speech",
+        params_b=0.082,
+        runtime="local / provider",
+        license="apache-2.0",
+        source_url="https://hf.co/hexgrad/Kokoro-82M",
+    ),
+    ModelCandidate(
+        repo_id="ResembleAI/chatterbox",
+        role="audio_lore_tts_optional",
+        task="text-to-speech",
+        params_b=0.5,
+        runtime="provider / Modal",
+        license="mit",
+        source_url="https://hf.co/ResembleAI/chatterbox",
+    ),
+    ModelCandidate(
+        repo_id="netflix/void-model",
+        role="modal_video_repair",
+        task="video-to-video",
+        params_b=5.0,
+        runtime="Modal / 40GB+ VRAM",
+        license="apache-2.0",
+        public_demo=False,
+        source_url="https://hf.co/netflix/void-model",
+    ),
     ModelCandidate(
         repo_id="Wan-AI/Wan2.2-I2V-A14B-Diffusers",
         role="video_swap_preset",
@@ -169,24 +195,30 @@ ADAPTER_CATALOG: list[AdapterRecipe] = [
     ),
 ]
 
-DEFAULT_ACTIVE_STACK = [
-    "black-forest-labs/FLUX.2-klein-4B",
+RAVEN_QUALITY_STACK = [
+    "black-forest-labs/FLUX.2-klein-9B",
+    "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
     "nvidia/LocateAnything-3B",
     "openbmb/MiniCPM-V-4.6",
     "nvidia/NVIDIA-Nemotron-Parse-v1.2",
     "openbmb/MiniCPM5-1B",
     "onnx-community/functiongemma-270m-it-ONNX",
+    "hexgrad/Kokoro-82M",
 ]
 
-PRIVATE_RESEARCH_STACK = [
-    "black-forest-labs/FLUX.2-klein-9B",
-    "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
+TINY_TITAN_STACK = [
+    "black-forest-labs/FLUX.2-klein-4B",
     "nvidia/LocateAnything-3B",
     "openbmb/MiniCPM-V-4.6",
     "nvidia/NVIDIA-Nemotron-Parse-v1.2",
     "openbmb/MiniCPM5-1B",
+    "onnx-community/functiongemma-270m-it-ONNX",
+    "hexgrad/Kokoro-82M",
 ]
 
+DEFAULT_ACTIVE_STACK = RAVEN_QUALITY_STACK
+PRIVATE_RESEARCH_STACK = RAVEN_QUALITY_STACK
+
 
 def filter_catalog(adult_mode: bool = False) -> tuple[list[ModelCandidate], list[AdapterRecipe]]:
     models = [
@@ -201,8 +233,12 @@ def filter_catalog(adult_mode: bool = False) -> tuple[list[ModelCandidate], list
 def active_stack(adult_mode: bool = False) -> list[ModelCandidate]:
     allowed, _ = filter_catalog(adult_mode)
     by_id = {model.repo_id: model for model in allowed}
-    stack_ids = PRIVATE_RESEARCH_STACK if adult_mode else DEFAULT_ACTIVE_STACK
-    return [by_id[repo_id] for repo_id in stack_ids if repo_id in by_id]
+    return [by_id[repo_id] for repo_id in RAVEN_QUALITY_STACK if repo_id in by_id]
+
+
+def tiny_titan_stack() -> list[ModelCandidate]:
+    by_id = {model.repo_id: model for model in MODEL_CATALOG}
+    return [by_id[repo_id] for repo_id in TINY_TITAN_STACK if repo_id in by_id]
 
 
 def parameter_budget(stack: list[ModelCandidate] | None = None) -> dict[str, float | str]:
@@ -223,5 +259,7 @@ def catalog_summary(adult_mode: bool = False) -> dict[str, int | float | str]:
         "models_visible": len(models),
         "adapters_visible": len(adapters),
         "adult_catalog": "enabled" if adult_mode else "hidden",
+        "active_preset": "Raven Quality Stack",
+        "tiny_titan": "sidecar",
         **budget,
     }

src/nexus_visual_weaver/exporter.py

@@ -84,18 +84,42 @@ def write_export_packet(
     artifact = _artifact_name(generation.get("output_path"))
     if "output_path" in generation:
         generation["output_path"] = artifact
+    modal_job = operator_state.get("modal_video_repair") or {
+        "status": "deferred",
+        "repo_id": "netflix/void-model",
+        "provider": "modal",
+    }
+    audio_lore = operator_state.get("audio_lore_tts") or {
+        "status": "optional",
+        "repo_id": "hexgrad/Kokoro-82M",
+    }
+    offellia = operator_state.get("offellia_judge") or {
+        "status": "deferred_local",
+        "repo_id": "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf",
+    }
+    tiny_titan = operator_state.get("tiny_titan_sidecar") or {
+        "status": "available",
+        "repo_id": "black-forest-labs/FLUX.2-klein-4B",
+    }
+    locate_grounding = operator_state.get("locateanything_grounding") or {}
     packet = {
         "schema": "nexus_visual_weaver.export_packet.v1",
         "run_id": run_id,
         "created_at_epoch": int(time.time()),
+        "active_preset": operator_state.get("active_preset", "Raven Quality Stack"),
         "adult_mode": run_adult_mode,
         "prompt": getattr(getattr(run, "request", None), "prompt", ""),
         "refined_prompt": getattr(getattr(run, "refined_prompt", None), "refined", ""),
         "artifact": artifact,
         "generation": generation,
         "st3gg_scan": scan,
+        "locateanything_grounding": locate_grounding,
+        "offellia_judge": offellia,
         "minicpm_judge": operator_state.get("minicpm_judge") or {},
         "nemotron_evidence": operator_state.get("nemotron_evidence") or {},
+        "modal_video_repair": modal_job,
+        "audio_lore_tts": audio_lore,
+        "tiny_titan_sidecar": tiny_titan,
         "checkpoint": {
             "status": operator_state.get("checkpoint"),
             "message": operator_state.get("message"),
@@ -118,6 +142,11 @@ def write_export_packet(
             "off_brand_custom_ui": True,
             "openbmb_lane": (operator_state.get("minicpm_judge") or {}).get("status") == "success",
             "nvidia_nemotron_lane": (operator_state.get("nemotron_evidence") or {}).get("status") == "success",
+            "offellia_quality_lane": offellia.get("status") in {"success", "completed"},
+            "modal_void_lane": modal_job.get("status") in {"success", "completed", "documented"},
+            "tiny_titan_sidecar": tiny_titan.get("status") in {"success", "available", "sidecar"},
+            "raven_quality_stack": True,
+            "locateanything_grounding": bool(locate_grounding.get("targets") or locate_grounding.get("repo_id")),
             "st3gg_export_gate": scan.get("export_gate"),
         },
     }

src/nexus_visual_weaver/hf_runtime.py

@@ -9,8 +9,9 @@ from pathlib import Path
 from typing import Any
 
 
-FLUX_REPO_ID = "black-forest-labs/FLUX.2-klein-4B"
-PRIVATE_RESEARCH_FLUX_REPO_ID = "black-forest-labs/FLUX.2-klein-9B"
+FLUX_REPO_ID = "black-forest-labs/FLUX.2-klein-9B"
+TINY_TITAN_FLUX_REPO_ID = "black-forest-labs/FLUX.2-klein-4B"
+PRIVATE_RESEARCH_FLUX_REPO_ID = FLUX_REPO_ID
 
 
 @dataclass(frozen=True)
@@ -65,13 +66,23 @@ def _hf_token() -> str | None:
     return os.environ.get("HF_TOKEN") or os.environ.get("HUGGING_FACE_HUB_TOKEN")
 
 
+def active_flux_repo_id() -> str:
+    configured = os.environ.get("NEXUS_FLUX_REPO_ID")
+    if configured:
+        return configured
+    if os.environ.get("NEXUS_TINY_TITAN_MODE") == "1":
+        return TINY_TITAN_FLUX_REPO_ID
+    return FLUX_REPO_ID
+
+
 def generate_flux_image(prompt: str, *, seed: int = 0, width: int = 1024, height: int = 1024, steps: int = 4) -> HFGenerationResult:
+    repo_id = active_flux_repo_id()
     if not hf_runtime_enabled():
         return HFGenerationResult(
             status="disabled",
             provider_state="dry-run",
-            repo_id=FLUX_REPO_ID,
-            message="Real HF generation disabled outside Space. Set NEXUS_ENABLE_REAL_HF=1 to force local execution.",
+            repo_id=repo_id,
+            message="Real HF generation disabled outside Space. Raven Quality Stack uses FLUX.2 Klein 9B by default; set NEXUS_TINY_TITAN_MODE=1 for the 4B sidecar.",
             width=width,
             height=height,
             steps=steps,
@@ -86,7 +97,7 @@ def generate_flux_image(prompt: str, *, seed: int = 0, width: int = 1024, height
         return HFGenerationResult(
             status="missing_runtime",
             provider_state="blocked",
-            repo_id=FLUX_REPO_ID,
+            repo_id=repo_id,
             message=f"FLUX runtime import failed. Install diffusers main + torch. {_short_error(exc)}",
             width=width,
             height=height,
@@ -98,51 +109,63 @@ def generate_flux_image(prompt: str, *, seed: int = 0, width: int = 1024, height
         return HFGenerationResult(
             status="no_cuda",
             provider_state="blocked",
-            repo_id=FLUX_REPO_ID,
-            message="CUDA is not available to the Space callback; FLUX.2 Klein 4B requires GPU execution.",
+            repo_id=repo_id,
+            message="CUDA is not available to the Space callback; FLUX.2 generation requires GPU execution.",
             width=width,
             height=height,
             steps=steps,
             hf_token_present=bool(_hf_token()),
         )
 
-    try:
-        dtype = torch.bfloat16
-        token = _hf_token()
-        pipe = FluxPipeline.from_pretrained(FLUX_REPO_ID, torch_dtype=dtype, token=token)
-        pipe.enable_model_cpu_offload()
-        generator = torch.Generator(device="cuda").manual_seed(seed)
-        image = pipe(
-            prompt=prompt,
-            height=height,
-            width=width,
-            guidance_scale=1.0,
-            num_inference_steps=steps,
-            generator=generator,
-        ).images[0]
-        output_path = _output_dir() / f"nexus_flux_{int(time.time())}_{seed}.png"
-        image.save(output_path)
-        return HFGenerationResult(
-            status="success",
-            provider_state="generated",
-            repo_id=FLUX_REPO_ID,
-            output_path=str(output_path),
-            message="FLUX.2 Klein 4B generated a real public-demo artifact on HF Space.",
-            latency_seconds=round(time.perf_counter() - started, 2),
-            width=width,
-            height=height,
-            steps=steps,
-            hf_token_present=bool(token),
-        )
-    except Exception as exc:  # pragma: no cover - exercised on HF Space with gated/runtime conditions.
-        return HFGenerationResult(
-            status="error",
-            provider_state="blocked",
-            repo_id=FLUX_REPO_ID,
-            message=f"FLUX.2 generation failed. Check model license acceptance, HF_TOKEN/Space access, and runtime deps. {_short_error(exc)}",
-            latency_seconds=round(time.perf_counter() - started, 2),
-            width=width,
-            height=height,
-            steps=steps,
-            hf_token_present=bool(_hf_token()),
-        )
+    token = _hf_token()
+    repo_candidates = [repo_id]
+    if repo_id != TINY_TITAN_FLUX_REPO_ID and os.environ.get("NEXUS_DISABLE_TINY_TITAN_FALLBACK") != "1":
+        repo_candidates.append(TINY_TITAN_FLUX_REPO_ID)
+    errors: list[str] = []
+    for candidate in repo_candidates:
+        try:
+            dtype = torch.bfloat16
+            pipe = FluxPipeline.from_pretrained(candidate, torch_dtype=dtype, token=token)
+            pipe.enable_model_cpu_offload()
+            generator = torch.Generator(device="cuda").manual_seed(seed)
+            image = pipe(
+                prompt=prompt,
+                height=height,
+                width=width,
+                guidance_scale=1.0,
+                num_inference_steps=steps,
+                generator=generator,
+            ).images[0]
+            output_path = _output_dir() / f"nexus_flux_{int(time.time())}_{seed}.png"
+            image.save(output_path)
+            fallback = candidate != repo_id
+            message = (
+                f"{candidate} generated a Tiny Titan sidecar artifact after the 9B lane was unavailable."
+                if fallback
+                else f"{candidate} generated a real Raven Quality artifact on HF Space."
+            )
+            return HFGenerationResult(
+                status="success",
+                provider_state="generated",
+                repo_id=candidate,
+                output_path=str(output_path),
+                message=message,
+                latency_seconds=round(time.perf_counter() - started, 2),
+                width=width,
+                height=height,
+                steps=steps,
+                hf_token_present=bool(token),
+            )
+        except Exception as exc:  # pragma: no cover - exercised on HF Space with gated/runtime conditions.
+            errors.append(f"{candidate}: {_short_error(exc)}")
+    return HFGenerationResult(
+        status="error",
+        provider_state="blocked",
+        repo_id=repo_id,
+        message=f"FLUX.2 generation failed. Check model license acceptance, HF_TOKEN/Space access, and runtime deps. Attempts: {' | '.join(errors)}",
+        latency_seconds=round(time.perf_counter() - started, 2),
+        width=width,
+        height=height,
+        steps=steps,
+        hf_token_present=bool(token),
+    )

src/nexus_visual_weaver/model_relay.py

@@ -15,6 +15,7 @@ from typing import Any
 
 PINNED_LANES = {"image_generation", "grounding", "security"}
 ROTATABLE_LANES = {
+    "tiny_titan_sidecar",
     "private_image_research",
     "prompt_router",
     "taste_judge",
@@ -380,7 +381,7 @@ class WeaverModelRelay:
 
     def _decision_reason(self, primary: ModelRecord, strategy: str, public_demo: bool) -> str:
         if strategy == "license_safe_public":
-            return f"{primary.model_id} selected because it is public-demo safe and within helper budget."
+            return f"{primary.model_id} selected because it is license-safe and within helper budget."
         if strategy == "quota_saver":
             return f"{primary.model_id} selected to preserve provider quota and reuse cheaper metadata paths."
         if strategy == "latency_first":
@@ -431,18 +432,32 @@ class WeaverModelRelay:
 def default_model_records() -> list[ModelRecord]:
     return [
         ModelRecord(
-            model_id="flux2-klein-4b-public",
+            model_id="flux2-klein-9b-quality",
             lane="image_generation",
             provider="hf",
+            repo_id="black-forest-labs/FLUX.2-klein-9B",
+            license_gate="review_required",
+            params_b=9.0,
+            cost_hint="gated_provider_or_quality_space",
+            rpm_limit=6,
+            rpd_limit=40,
+            quality_score=0.97,
+            latency_ms=26000,
+            pinned=True,
+        ),
+        ModelRecord(
+            model_id="flux2-klein-4b-tiny-sidecar",
+            lane="tiny_titan_sidecar",
+            provider="hf",
             repo_id="black-forest-labs/FLUX.2-klein-4B",
             license_gate="apache-2.0",
             params_b=4.0,
-            cost_hint="provider_or_local",
+            cost_hint="public_fallback_or_tiny_titan_export",
             rpm_limit=8,
             rpd_limit=60,
             quality_score=0.92,
             latency_ms=21000,
-            pinned=True,
+            fallback_chain=("flux2-klein-9b-quality",),
         ),
         ModelRecord(
             model_id="flux2-klein-9b-private",
@@ -451,11 +466,12 @@ def default_model_records() -> list[ModelRecord]:
             repo_id="black-forest-labs/FLUX.2-klein-9B",
             license_gate="review_required",
             params_b=9.0,
-            cost_hint="gated_provider_or_private_space",
+            cost_hint="legacy_quality_alias",
             rpm_limit=6,
             rpd_limit=40,
             quality_score=0.96,
             latency_ms=26000,
+            health="excluded",
         ),
         ModelRecord(
             model_id="locateanything-3b-anchor",
@@ -758,13 +774,13 @@ def default_model_records() -> list[ModelRecord]:
             model_id="netflix-void-modal",
             lane="video_repair",
             provider="modal",
-            repo_id="Netflix/VOID",
-            license_gate="private_research",
-            params_b=1.3,
-            cost_hint="modal_credits",
-            rpm_limit=10,
-            rpd_limit=120,
-            quality_score=0.84,
+            repo_id="netflix/void-model",
+            license_gate="apache-2.0",
+            params_b=5.0,
+            cost_hint="modal_credits_40gb_vram",
+            rpm_limit=4,
+            rpd_limit=30,
+            quality_score=0.88,
             latency_ms=12000,
             fallback_chain=("void-q5-offline",),
         ),
@@ -772,9 +788,9 @@ def default_model_records() -> list[ModelRecord]:
             model_id="void-q5-offline",
             lane="video_repair",
             provider="local",
-            repo_id="local/VOID-Q5-video-repair",
+            repo_id="local/netflix-void-q5-video-repair",
             license_gate="private_research",
-            params_b=1.3,
+            params_b=5.0,
             cost_hint="offline",
             rpm_limit=20,
             rpd_limit=200,

src/nexus_visual_weaver/render.py

@@ -113,12 +113,13 @@ def render_command_header() -> str:
       <div>
         <small>COMMAND INPUT</small>
         <strong>Raven Chronicle Active Weave</strong>
-        <span>Prompt, reference scan, model route, and checkpoint controls stay in one sticky operator strip.</span>
+        <span>Quality stack, reference grounding, model evidence, and checkpoint controls stay in one sticky operator strip.</span>
       </div>
       <div class="nw-command-pills">
-        {badge("SFW DEFAULT", "pass")}
+        {badge("RAVEN QUALITY STACK", "pass")}
         {badge("ST3GG ALWAYS ON", "cyan")}
-        {badge("FLUX.2 4B PINNED", "accent")}
+        {badge("FLUX.2 9B PINNED", "accent")}
+        {badge("4B TINY SIDECAR", "muted")}
         {badge("HUMAN CHECKPOINT", "warn")}
       </div>
     </section>
@@ -181,7 +182,7 @@ def render_topbar(
     <div class="nw-topbar">
       <div class="nw-brand"><span>NEXUS</span><strong>Visual Weaver</strong></div>
       <div class="nw-topitem"><small>Project</small><strong>Raven Chronicle</strong><i></i></div>
-      <div class="nw-topitem"><small>Active Preset</small><strong>Dark Couture v2.4</strong><i></i></div>
+      <div class="nw-topitem"><small>Active Preset</small><strong>Raven Quality Stack</strong><i></i></div>
       <div class="nw-budget">
         <div><strong>32B Parameter Budget</strong><small>{active:.2f}B / 32B ({pct}%)</small></div>
         <div class="nw-meter"><i style="width:{pct}%"></i></div>
@@ -256,7 +257,7 @@ def render_workflow(run: GenerationRun | None = None, operator_state: dict | Non
         "refine": (275, 52, 185, 160, "Refine", ["Prompt Refiner", "Style Harmonizer", "Negative Purge"], "Qwen2.5-7B", "complete", "violet"),
         "judge": (540, 52, 185, 160, "Judge", ["Aesthetic Scorer", "ST3GG Policy Filter", f"Score {score:.2f}"], "MiniCPM / Nemotron", "complete", "blue"),
         "locate": (785, 52, 185, 160, "Locate", ["Reference Locator", "Pose & Composition", "IP-Adapter"], "Refs 3/5", "complete", "cyan"),
-        "generate": (275, 280, 235, 210, "Generate", ["Image / Video Generation", "FLUX.2 4B + adapter stack", "High-detail couture"], "Steps 4  CFG 1.0", "ready", "green"),
+        "generate": (275, 280, 235, 210, "Generate", ["Image / Video Generation", "FLUX.2 9B quality lane", "High-detail couture"], "Steps 4  CFG 1.0", "ready", "green"),
         "video": (590, 280, 235, 210, "Video Path", ["Image to Video", "Frame interpolation", run.video.preset if run else "Wan2.2 / LTX swap"], "Duration 5.6s  24fps", "ready", "blue"),
         "checkpoint": (880, 285, 185, 185, "Human Checkpoint", ["Human review required", "Verify intent, vibe,", "and output before final."], "Review Now", "paused", "amber"),
     }
@@ -569,7 +570,7 @@ def _render_relay_panel(relay_status: dict | None = None) -> str:
     <h3>GMR ModelRelay</h3>
     <ul class="nw-relay">{rows}</ul>
     <div class="nw-relay-foot">
-      {badge("FLUX.2 4B pinned", "pass")} {badge("LocateAnything pinned", "pass")} {badge(f"dedup hits {dedup_hits}", "muted")}
+      {badge("FLUX.2 9B pinned", "pass")} {badge("4B sidecar", "muted")} {badge("LocateAnything pinned", "pass")} {badge(f"dedup hits {dedup_hits}", "muted")}
     </div>
     """
 
@@ -590,6 +591,7 @@ def render_provider_cards(relay_status: dict | None = None, adult_mode: bool = F
     optional_statuses = {
         "openbmb": "configured" if _provider_configured("MINICPM_BASE_URL", "MINICPM_API_KEY", "OPENBMB_API_KEY") else "missing secret",
         "nvidia": "configured" if _provider_configured("NEMOTRON_BASE_URL", "NEMOTRON_API_KEY", "NVIDIA_API_KEY") else "missing secret",
+        "modal": "configured" if _env_configured("MODAL_TOKEN_ID", "MODAL_TOKEN_SECRET") else "deferred",
         "fal": "configured" if _env_configured("FAL_KEY") else "blocked",
         "netlify": "configured" if _env_configured("NETLIFY_AUTH_TOKEN", "NETLIFY_SITE_ID", "OPENAI_BASE_URL") else "blocked",
         "cloudflare": "configured" if _env_configured("CLOUDFLARE_API_TOKEN", "CF_ACCOUNT_ID") else "blocked",
@@ -626,9 +628,9 @@ def render_provider_cards(relay_status: dict | None = None, adult_mode: bool = F
             <div class="nw-provider-card nw-provider-optional">
               <small>optional gateway</small>
               <strong>{escape(provider.title())}</strong>
-              <span>off by default / secrets required</span>
-              <i class="nw-provider-meter" style="--health:{'74' if state == 'configured' else '18'}"></i>
-              <div>{badge(state.upper(), "pass" if state == "configured" else "warn")}{badge("SPONSOR LANE" if provider in {"openbmb", "nvidia", "hf_nvidia"} else "NOT MVP DEFAULT", "muted")}</div>
+              <span>{"VOID repair job / Modal credit lane" if provider == "modal" else "off by default / secrets required"}</span>
+              <i class="nw-provider-meter" style="--health:{'74' if state == 'configured' else '42' if state == 'deferred' else '18'}"></i>
+              <div>{badge(state.upper(), "pass" if state == "configured" else "muted" if state == "deferred" else "warn")}{badge("SPONSOR LANE" if provider in {"openbmb", "nvidia", "hf_nvidia", "modal"} else "NOT MVP DEFAULT", "muted")}</div>
             </div>
             """
         )
@@ -673,7 +675,7 @@ def render_inspector(
         scan_status = (scan or {}).get("status", "pass")
     else:
         checks = [(label, True) for label in ["Patent Leather", "Faux Fur", "Lace / Mesh", "Crimson Hardware", "Platform Boots", "Layered Garments"]]
-        model_rows = "<li><span>active stack</span><strong>FLUX.2 4B / MiniCPM / LocateAnything</strong></li>"
+        model_rows = "<li><span>active stack</span><strong>FLUX.2 9B / OFFELLIA / LocateAnything</strong></li>"
         score = 86
         scan_status = (scan or {}).get("status", "pass")
     checks_html = "".join(f'<li><span>{"✓" if ok else "!"}</span>{escape(label)}</li>' for label, ok in checks)
@@ -682,11 +684,19 @@ def render_inspector(
     operator_state = operator_state or {}
     minicpm = operator_state.get("minicpm_judge") or {}
     nemotron = operator_state.get("nemotron_evidence") or {}
+    offellia = operator_state.get("offellia_judge") or {}
+    modal = operator_state.get("modal_video_repair") or {}
+    tts = operator_state.get("audio_lore_tts") or {}
+    locate = operator_state.get("locateanything_grounding") or {}
     sponsor_rows = "".join(
         f"<li><span>{escape(label)}</span><strong>{escape(str(result.get('status', 'pending')).upper())}</strong><em>{escape(str(result.get('repo_id', repo)))}</em></li>"
         for label, repo, result in [
+            ("OFFELLIA Quality Judge", "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf", offellia),
             ("OpenBMB MiniCPM", "openbmb/MiniCPM-V-4.6", minicpm),
             ("NVIDIA Nemotron", "nvidia/NVIDIA-Nemotron-Parse-v1.2", nemotron),
+            ("LocateAnything Grounding", "nvidia/LocateAnything-3B", locate),
+            ("Modal VOID Repair", "netflix/void-model", modal),
+            ("Audio Lore TTS", "hexgrad/Kokoro-82M", tts),
         ]
     )
     findings = [_redact_scan_text(item) for item in (scan.get("findings") or [])]

tests/test_command_center.py

@@ -3,10 +3,13 @@ from pathlib import Path
 from nexus_visual_weaver.catalog import (
     DEFAULT_ACTIVE_STACK,
     PRIVATE_RESEARCH_STACK,
+    RAVEN_QUALITY_STACK,
+    TINY_TITAN_STACK,
     active_stack,
     catalog_summary,
     filter_catalog,
     parameter_budget,
+    tiny_titan_stack,
 )
 from nexus_visual_weaver.grounding import inspect_outfit
 from nexus_visual_weaver.model_relay import WeaverModelRelay
@@ -63,13 +66,22 @@ def test_active_parameter_budget_passes_default_stack() -> None:
     assert budget["active_b"] <= 32.0
 
 
-def test_public_stack_uses_flux_4b_and_excludes_private_models() -> None:
+def test_default_stack_uses_raven_quality_models() -> None:
     stack = active_stack(False)
     repo_ids = {model.repo_id for model in stack}
 
+    assert "black-forest-labs/FLUX.2-klein-9B" in repo_ids
+    assert "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf" in repo_ids
+    assert "black-forest-labs/FLUX.2-klein-4B" not in repo_ids
+    assert all(model.params_b < 32.0 for model in stack)
+
+
+def test_tiny_titan_stack_is_sidecar_only_and_all_models_are_under_4b() -> None:
+    stack = tiny_titan_stack()
+    repo_ids = {model.repo_id for model in stack}
+
     assert "black-forest-labs/FLUX.2-klein-4B" in repo_ids
     assert "black-forest-labs/FLUX.2-klein-9B" not in repo_ids
-    assert not any("OFFELLIA" in repo_id for repo_id in repo_ids)
     assert all(model.params_b <= 4.0 for model in stack)
 
 
@@ -213,7 +225,8 @@ def test_command_header_exposes_governed_run_controls() -> None:
 
     assert "Raven Chronicle Active Weave" in header
     assert "ST3GG ALWAYS ON" in header
-    assert "FLUX.2 4B PINNED" in header
+    assert "FLUX.2 9B PINNED" in header
+    assert "4B TINY SIDECAR" in header
     assert "HUMAN CHECKPOINT" in header
 
 
@@ -359,7 +372,7 @@ def test_render_inspector_with_success_judge_shows_success_status() -> None:
 def test_render_inspector_shows_default_stack_label_without_run() -> None:
     html = render_inspector()
 
-    assert "FLUX.2 4B / MiniCPM / LocateAnything" in html
+    assert "FLUX.2 9B / OFFELLIA / LocateAnything" in html
 
 
 # --- render_provider_cards sponsor lane tests ---
@@ -432,11 +445,11 @@ def test_render_operations_and_inspector_redact_payload_details() -> None:
 
 # --- catalog public_demo field tests ---
 
-def test_filter_catalog_excludes_flux_9b_in_public_mode() -> None:
+def test_filter_catalog_includes_flux_9b_in_public_mode() -> None:
     models, _ = filter_catalog(False)
     repo_ids = {model.repo_id for model in models}
 
-    assert "black-forest-labs/FLUX.2-klein-9B" not in repo_ids
+    assert "black-forest-labs/FLUX.2-klein-9B" in repo_ids
 
 
 def test_filter_catalog_includes_flux_9b_in_adult_mode() -> None:
@@ -446,11 +459,11 @@ def test_filter_catalog_includes_flux_9b_in_adult_mode() -> None:
     assert "black-forest-labs/FLUX.2-klein-9B" in repo_ids
 
 
-def test_filter_catalog_excludes_offellia_in_public_mode() -> None:
+def test_filter_catalog_includes_offellia_in_public_mode() -> None:
     models, _ = filter_catalog(False)
     repo_ids = {model.repo_id for model in models}
 
-    assert not any("OFFELLIA" in repo_id for repo_id in repo_ids)
+    assert "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf" in repo_ids
 
 
 def test_active_stack_uses_private_research_stack_in_adult_mode() -> None:
@@ -458,6 +471,7 @@ def test_active_stack_uses_private_research_stack_in_adult_mode() -> None:
     repo_ids = {model.repo_id for model in stack}
 
     assert "black-forest-labs/FLUX.2-klein-9B" in repo_ids
+    assert "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf" in repo_ids
     assert "black-forest-labs/FLUX.2-klein-4B" not in repo_ids
 
 
@@ -466,11 +480,14 @@ def test_private_research_stack_constant_contains_9b_and_offellia() -> None:
     assert "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf" in PRIVATE_RESEARCH_STACK
 
 
-def test_default_active_stack_constant_uses_4b_and_sponsor_models() -> None:
-    assert "black-forest-labs/FLUX.2-klein-4B" in DEFAULT_ACTIVE_STACK
+def test_default_active_stack_constant_uses_9b_and_sponsor_models() -> None:
+    assert DEFAULT_ACTIVE_STACK == RAVEN_QUALITY_STACK
+    assert "black-forest-labs/FLUX.2-klein-9B" in DEFAULT_ACTIVE_STACK
+    assert "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf" in DEFAULT_ACTIVE_STACK
     assert "openbmb/MiniCPM-V-4.6" in DEFAULT_ACTIVE_STACK
     assert "nvidia/NVIDIA-Nemotron-Parse-v1.2" in DEFAULT_ACTIVE_STACK
-    assert "black-forest-labs/FLUX.2-klein-9B" not in DEFAULT_ACTIVE_STACK
+    assert "black-forest-labs/FLUX.2-klein-4B" not in DEFAULT_ACTIVE_STACK
+    assert "black-forest-labs/FLUX.2-klein-4B" in TINY_TITAN_STACK
 
 
 # --- schema ModelCandidate public_demo tests ---
@@ -510,7 +527,8 @@ def test_public_demo_false_models_are_excluded_from_public_filter() -> None:
         license="other",
         public_demo=False,
     )
-    # public_demo=False should mean filter_catalog(False) excludes it
-    # The catalog-level test: verify FLUX 9B (public_demo=False) is absent
+    # public_demo=False should mean filter_catalog(False) excludes hidden support lanes.
+    # The catalog-level test: verify Modal VOID is absent from public catalog scope.
     models_public, _ = filter_catalog(False)
     assert all(m.public_demo for m in models_public)
+    assert "netflix/void-model" not in {model.repo_id for model in models_public}

tests/test_exporter.py

@@ -11,8 +11,13 @@ def _make_base_state(**overrides):
         "checkpoint": "approved",
         "message": "approved",
         "generation": {"status": "success", "output_path": "/data/artifact.png", "hf_token_present": True},
+        "locateanything_grounding": {"status": "pass", "repo_id": "nvidia/LocateAnything-3B", "targets": [{"slot_name": "footwear"}]},
+        "offellia_judge": {"status": "deferred_local", "repo_id": "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf"},
         "minicpm_judge": {"status": "success", "repo_id": "openbmb/MiniCPM-V-4.6"},
         "nemotron_evidence": {"status": "missing_secret", "repo_id": "nvidia/NVIDIA-Nemotron-Parse-v1.2"},
+        "modal_video_repair": {"status": "deferred", "repo_id": "netflix/void-model", "provider": "modal"},
+        "audio_lore_tts": {"status": "optional", "repo_id": "hexgrad/Kokoro-82M"},
+        "tiny_titan_sidecar": {"status": "available", "repo_id": "black-forest-labs/FLUX.2-klein-4B"},
     }
     state.update(overrides)
     return state
@@ -32,6 +37,14 @@ def test_write_export_packet_records_evidence_without_secrets(monkeypatch) -> No
     assert payload["hackathon_claims"]["openbmb_lane"] is True
     assert payload["hackathon_claims"]["nvidia_nemotron_lane"] is False
     assert payload["parameter_budget"]["status"] == "pass"
+    assert payload["active_preset"] == "Raven Quality Stack"
+    assert payload["modal_video_repair"]["repo_id"] == "netflix/void-model"
+    assert payload["offellia_judge"]["repo_id"] == "Brunobkr/OFFELLIA_Q4_0_gemma-4-12B-it.gguf"
+    assert payload["audio_lore_tts"]["repo_id"] == "hexgrad/Kokoro-82M"
+    assert payload["tiny_titan_sidecar"]["repo_id"] == "black-forest-labs/FLUX.2-klein-4B"
+    assert payload["hackathon_claims"]["raven_quality_stack"] is True
+    assert payload["hackathon_claims"]["locateanything_grounding"] is True
+    assert payload["hackathon_claims"]["offellia_quality_lane"] is False
     assert "token" not in json.dumps(payload).lower()
     assert payload["artifact"] == "artifact.png"
     assert payload["generation"]["output_path"] == "artifact.png"

tests/test_hf_runtime.py

@@ -1,6 +1,13 @@
 from PIL import Image
 
-from nexus_visual_weaver.hf_runtime import FLUX_REPO_ID, PRIVATE_RESEARCH_FLUX_REPO_ID, generate_flux_image, hf_runtime_enabled
+from nexus_visual_weaver.hf_runtime import (
+    FLUX_REPO_ID,
+    PRIVATE_RESEARCH_FLUX_REPO_ID,
+    TINY_TITAN_FLUX_REPO_ID,
+    active_flux_repo_id,
+    generate_flux_image,
+    hf_runtime_enabled,
+)
 from nexus_visual_weaver.render import render_artifact_lane
 
 
@@ -14,12 +21,20 @@ def test_hf_runtime_is_disabled_locally_by_default(monkeypatch) -> None:
     result = generate_flux_image("test prompt")
     assert result.status == "disabled"
     assert result.provider_state == "dry-run"
-    assert result.repo_id == "black-forest-labs/FLUX.2-klein-4B"
+    assert result.repo_id == "black-forest-labs/FLUX.2-klein-9B"
 
 
-def test_public_and_private_flux_repo_ids_are_split() -> None:
-    assert FLUX_REPO_ID == "black-forest-labs/FLUX.2-klein-4B"
+def test_quality_and_sidecar_flux_repo_ids_are_split(monkeypatch) -> None:
+    monkeypatch.delenv("NEXUS_FLUX_REPO_ID", raising=False)
+    monkeypatch.delenv("NEXUS_TINY_TITAN_MODE", raising=False)
+
+    assert FLUX_REPO_ID == "black-forest-labs/FLUX.2-klein-9B"
     assert PRIVATE_RESEARCH_FLUX_REPO_ID == "black-forest-labs/FLUX.2-klein-9B"
+    assert TINY_TITAN_FLUX_REPO_ID == "black-forest-labs/FLUX.2-klein-4B"
+    assert active_flux_repo_id() == FLUX_REPO_ID
+
+    monkeypatch.setenv("NEXUS_TINY_TITAN_MODE", "1")
+    assert active_flux_repo_id() == TINY_TITAN_FLUX_REPO_ID
 
 
 def test_artifact_lane_embeds_generated_image() -> None:

tests/test_model_relay.py

@@ -11,19 +11,20 @@ def test_pinned_lanes_do_not_rotate() -> None:
     assert decision.pinned is True
     assert decision.rotatable is False
     assert decision.primary is not None
-    assert decision.primary.repo_id == "black-forest-labs/FLUX.2-klein-4B"
-    assert decision.primary.params_b == 4.0
+    assert decision.primary.repo_id == "black-forest-labs/FLUX.2-klein-9B"
+    assert decision.primary.params_b == 9.0
     assert decision.fallbacks == []
     assert "rotation disabled" in decision.reason
 
 
-def test_private_image_research_keeps_flux_9b_available() -> None:
+def test_tiny_titan_sidecar_keeps_flux_4b_available() -> None:
     relay = WeaverModelRelay()
-    decision = relay.select_lane("private_image_research", budget=9.0, public_demo=False, strategy="private_research")
+    decision = relay.select_lane("tiny_titan_sidecar", budget=4.0, public_demo=True, strategy="license_safe_public")
 
     assert decision.primary is not None
-    assert decision.primary.repo_id == "black-forest-labs/FLUX.2-klein-9B"
+    assert decision.primary.repo_id == "black-forest-labs/FLUX.2-klein-4B"
     assert decision.primary.pinned is False
+    assert decision.primary.params_b == 4.0
 
 
 def test_public_private_taste_judge_respects_license_and_budget() -> None:
@@ -123,7 +124,8 @@ def test_dashboard_surfaces_gmr_pinned_models_and_fallbacks() -> None:
     html = render_dashboard(relay_status=relay.dashboard_snapshot(public_demo=True))
 
     assert "GMR ModelRelay" in html
-    assert "FLUX.2 4B pinned" in html
+    assert "FLUX.2 9B pinned" in html
+    assert "4B sidecar" in html
     assert "LocateAnything pinned" in html
     assert "fallback:" in html
     assert "Rotation Safe" in html
@@ -135,6 +137,8 @@ def test_optional_external_gateways_are_registered_but_excluded_by_default() ->
     assert relay.records["netlify-ai-gateway-helper"].provider == "netlify"
     assert relay.records["cloudflare-agent-helper"].provider == "cloudflare"
     assert relay.records["fal-media-adapter"].provider == "fal"
+    assert relay.records["netflix-void-modal"].repo_id == "netflix/void-model"
+    assert relay.records["netflix-void-modal"].params_b == 5.0
     assert relay.records["netflix-void-modal"].health == "healthy"
     assert relay.records["netlify-ai-gateway-helper"].health == "excluded"
     assert relay.records["fal-media-adapter"].health == "excluded"
@@ -184,27 +188,27 @@ def test_private_image_research_lane_is_rotatable() -> None:
     assert "private_image_research" in ROTATABLE_LANES
 
 
-def test_flux2_klein_4b_is_pinned_with_apache_license() -> None:
+def test_flux2_klein_9b_is_pinned_quality_lane() -> None:
     relay = WeaverModelRelay()
 
-    record = relay.records["flux2-klein-4b-public"]
+    record = relay.records["flux2-klein-9b-quality"]
 
     assert record.lane == "image_generation"
     assert record.pinned is True
-    assert record.params_b == 4.0
-    assert record.license_gate == "apache-2.0"
-    assert record.repo_id == "black-forest-labs/FLUX.2-klein-4B"
+    assert record.params_b == 9.0
+    assert record.license_gate == "review_required"
+    assert record.repo_id == "black-forest-labs/FLUX.2-klein-9B"
 
 
-def test_flux2_klein_9b_is_not_pinned_and_in_private_research() -> None:
+def test_flux2_klein_4b_is_tiny_titan_sidecar() -> None:
     relay = WeaverModelRelay()
 
-    record = relay.records["flux2-klein-9b-private"]
+    record = relay.records["flux2-klein-4b-tiny-sidecar"]
 
-    assert record.lane == "private_image_research"
+    assert record.lane == "tiny_titan_sidecar"
     assert record.pinned is False
-    assert record.params_b == 9.0
-    assert record.license_gate == "review_required"
+    assert record.params_b == 4.0
+    assert record.license_gate == "apache-2.0"
 
 
 def test_minicpm_has_fallback_chain_configured() -> None: