laion/Vivarium-Greenscreen-Characters-FLUX-9B

Vivarium Greenscreen Characters — FLUX 9B

This dataset contains fictional, illustrated foreground characters made for animation, games, storyboards and other creative projects. People are shown full-length against a solid green background so they can be separated from a scene and composited over a location image. The visual style is warm, hand-painted and gently anime-inspired; it is not photorealistic.

What you will find

The collection includes 500 people with ten facial expressions each, 240 additional adult people with ten expressions each, 100 fantasy or science-fiction characters, and a separate set of adult professional outfits. That means a person can be found by a stable personality_id; all expressions and outfits for that person should stay together when making train/test splits. The current public snapshot contains 8,810/12,140 images: 5,000/5,000 original images, 2,500/2,500 additional-expression and special-character images, and 1,310/4,640 professional-outfit images. The Hub's default preview starts with young- and middle-adult samples; the full dataset still includes the other age groups.

All people are fully clothed and presented in non-sexual, everyday poses. The collection includes children as well as adults; intended ages, heritage labels and other attributes describe how each image was prompted and are not verified identity or demographic measurements. See the published child/teen safety recheck for the detector policy and checksum-specific results.

Sample images

The first two examples show the same adult identity with different expressions. The other images show a professional outfit and one fictional fantasy character.

Adult, default expression	Same adult, joyful expression	Pharmacist outfit	Fantasy character

Choose a collection

The default configuration includes all available families. Use base for the original 500 people, humans_extra for the 240 additional adults, specials for fantasy and science-fiction subjects, or jobs for professional outfits. Images are 832 × 1,248 RGB PNGs stored inside TAR archives with a JSON record and prompt text beside each image.

from datasets import load_dataset

people = load_dataset("laion/Vivarium-Greenscreen-Characters-FLUX-9B", "base", split="train", streaming=True)
sample = next(iter(people))
print(sample["json"]["personality_id"], sample["json"]["age_group"], sample["json"]["emotion"])

The JSON record carries the prompt, stable identity, expression, intended age group, outfit, seed, model settings and image checksum. See the detailed age and character extension guide, professional outfit guide, complete original collection, and inference instructions. Images use the undistilled FLUX.2 Klein Base 9B with the Vivarium character LoRA, 50 steps, real CFG 4 and BF16. The companion location background dataset contains matching wide-format scenes.

Original 500-person collection details

500 fictional people ×10 natural facial expressions =5,000 portrait images, each832×1,248 pixels. The selected aesthetic is the ghibli-warm-painted direction: restrained facial proportions, delicate drawn lines, warm painted shading and a solid chroma-green backdrop. Generation uses the undistilled9B base, the rank64 character LoRA, 50 steps, real CFG4, and BF16, with no quantization or distilled checkpoint.

Complete: 500 people and 5,000 images generated, validated and uploaded.

The base model and character LoRA with examples and instructions are linked separately. The compatible background dataset contains wide warm-painted environments for these foreground characters.

People, ages and intended distribution

Age group	Intended ages	Male	Female	People	Images
baby	9–24 months sampled	22	22	44	440
preschool	3–5 years	22	22	44	440
younger_schoolchild	6–8 years	22	22	44	440
older_schoolchild	9–12 years	24	24	48	480
younger_teen	13–15 years	24	24	48	480
older_teen	16–18 years	24	24	48	480
young_adult	19–29 years	24	24	48	480
middle_adult	30–49 years	22	22	44	440
older_adult	50–64 years	22	22	44	440
younger_senior	65–79 years	22	22	44	440
older_senior	80–95 years	22	22	44	440

There are 250 male and 250 female prompt labels. Older schoolchildren, younger teens, older teens and young adults receive 24 people per gender; the other age groups receive 22. Each age/gender cell is exactly 50% white European-ancestry in the prompt; its remaining people rotate across seven other groups. Overall allocation is:

Intended prompt label	People	Images
white_caucasian	250	2500
african_american	36	360
arab	36	360
latino	36	360
indian	36	360
japanese	36	360
southeast_asian	35	350
indigenous_american	35	350

These are intended prompt attributes, not ethnicity or gender inferred from rendered appearance, biological categories, or ground-truth measurements. Skin tones, hair, clothing and builds vary; no cultural costume is required to indicate a group. The distribution is deliberately selected coverage rather than population prevalence. Infants are sampled at 9, 12, 15, 18, 21 or 24 months and safely seated with the full clothed body visible. Other ages use a standing head-to-thigh composition; the model often renders the full body instead. Teen prompts explicitly request natural adolescent proportions; late teens use near-adult faces with modest eyes and a defined jaw, while schoolchildren retain age-appropriate proportions without oversized doll-like eyes.

The full class list, taxonomy, exact allocation, 500-person CSV, and full prompt plan expose every attribute. No real people's photographs or external audio references were used.

Stable person IDs and expression variants

personality_id and character_id are identical stable keys such as person-0253. A person has one outfit, identified by outfit_id=outfit-01, and ten sample IDs such as person-0253-neutral and person-0253-angry. The neutral/slightly friendly image is the default. This release does not generate outfit changes; later outfits can retain the person ID and add new outfit IDs.

Expression ID	Intended mild facial cue	Images
neutral	a relaxed neutral expression with a slight friendly smile	500
joyful	natural joy: an open smiling mouth, raised cheeks and bright happy eyes	500
grateful	warm gratitude: a gentle closed-mouth appreciative smile and softly relaxed eyelids	500
playful	gentle playfulness: a small lopsided mischievous grin and one eyebrow slightly raised	500
angry	mild controlled anger: eyebrows drawn inward and lowered, lips firmly pressed together	500
afraid	mild apprehension: raised inner eyebrows, slightly widened eyes and gently parted lips	500
sad	quiet mild sadness: raised inner eyebrows, softly sad eyes and downturned mouth corners	500
annoyed	mild annoyance: partly lowered eyelids, one brow lowered and a faint flat line of the lips	500
pain	brief mild discomfort: a slight wince, gently narrowed eyes and compressed lips, with no visible injury	500
disgust	mild distaste: a slight nose wrinkle, subtly raised upper lip and asymmetric downturned mouth	500

All expressions request mild natural facial changes, with clearly specified mouth, brow and eyelid cues, rather than exaggerated caricatures. Joyful uses a natural open smile; grateful uses a softer appreciative smile. Pain means mild discomfort with no visible injury. Playful means an innocent lopsided smile. Fully clothed age-appropriate everyday outfits are used for all ages, including children and adolescents. Outfit styles range across casual layers, sportswear, formal clothing, practical workwear and rock-inspired denim; clothing avoids green and teal to support later chroma-keying.

Clothing styles and colors use independent identity-hashed allocations, so the even/odd demographic assignment does not determine a clothing family. An early allocation refinement preserved the 24 already started people and reassigned only the 476 queued people; the audit records those changes. The published full person plan is authoritative for reproduction, including those retained initial outfits.

The neutral image is generated first. Each of the nine other images uses that same neutral image as its visual reference, resized to 416×624 using PIL BICUBIC. Original groups use the same seed throughout; detector-driven replacements can retry with separate per-expression seeds, recorded exactly in the plan and each PNG's JSON. Edits request the same identity, hair, skin, outfit, pose and framing and modify the face. The full-size neutral source PNG's SHA256 and sample ID are recorded; the resize parameters identify how to reconstruct the actual reference input. Reference conditioning encourages consistency; it is not a pixel-locking identity guarantee. Small pose, facial-detail or clothing changes can occur. The visual review records what was inspected rather than claiming all appearances were automatically verified. See the checksum-specific safety recheck.

An initial pilot preserved identity but made facial expressions too similar. It was rejected before public ingestion. Subsequent pilots use focused face-edit prompts instead of repeating the full person description and remove the generic expression-suppression negative clause from edits. The accepted quality report, when available, records the actual inspected samples.

A later visual check found lighter facial skin in one senior character's joyful edit. That complete ten-image group was removed and regenerated with explicit preservation of the original skin tone, exposure and lighting. The revised wording improved the inspected series and was applied to the 453 people still queued at that point. Already started groups retained their actual prompts. The targeted regeneration and prompt refinement audit record the changes. Small skin-tone or lighting variations can still occur; the full plan and each sample's metadata contain the wording actually used.

Prompt template and example

The shared style is:

Studio Ghibli-inspired grounded coming-of-age animation, delicate expressive hand-drawn lines, rich warm watercolor-like painted shading, gentle luminous colors, restrained almond-shaped eyes and proportionate faces.

For a default portrait, tools/plan_character_production.py builds a complete age/heritage/hair/outfit description and applies the shared character template. For edits it uses a focused instruction, for example:

vvar_char, edit the reference portrait. Make this exact same fictional 75-year-old person show mild controlled anger: eyebrows drawn inward and lowered, lips firmly pressed together. Change the mouth, eyebrows and eyelids so this mild natural emotion is clearly visible. Preserve exact identity, face shape, skin tone, hair, clothing, pose and framing. Keep the original deep brown skin tone exactly, with identical exposure and lighting. Studio Ghibli-inspired grounded coming-of-age animation, delicate expressive hand-drawn lines, rich warm watercolor-like painted shading, gentle luminous colors, restrained almond-shaped eyes and proportionate faces. Fully clothed, age-appropriate. Perfectly uniform solid bright chroma green background. No scenery, no text, no logo, no watermark.

Negative prompts exclude extra people, changed clothing, different hairstyles, text, logos and scenery. School-age and older prompts also discourage oversized doll eyes and chibi proportions. The full positive and negative prompts are in every sample's JSON and the actual positive prompt also has its own TXT member.

WebDataset layout and loading

Every sample contains three adjacent members:

person-0253-neutral.png
person-0253-neutral.json
person-0253-neutral.txt

Images are written directly into TARs. There are no individually uploaded PNGs. A production shard normally holds two complete person groups, 20 images, with 10-image first-worker/pilot shards and smaller last-worker shards. Keeping all expressions of a person together makes grouping convenient. Shard SHA256/size/count manifest, inference records, progress, and sample index describe the release snapshot. Use personality_id for splitting: random splitting by image can leak the same fictional identity across train and evaluation.

Install the lightweight reader dependencies and stream with Hugging Face Datasets:

from datasets import load_dataset
ds = load_dataset("laion/Vivarium-Greenscreen-Characters-FLUX-9B",
                  split="train", streaming=True)
sample = next(iter(ds))
image, metadata, prompt = sample["png"], sample["json"], sample["txt"]
print(metadata["personality_id"], metadata["emotion"], image.size)

The base configuration selects only data/*.tar; the default also includes available extension TAR families. JSON indexes are excluded from training examples. An alternate standard-library TAR/HTTP reader avoids extracting image files:

python tools/read_webdataset.py pilot-person-0253.tar \
  --repo laion/Vivarium-Greenscreen-Characters-FLUX-9B --limit 10

Inference and reproduction

Install requirements.txt, obtain access to the linked base model, and authenticate locally using your own Hugging Face account. The generator uses pinned model and LoRA revisions listed in provenance; no access token is included in the dataset or code. The default example generates one late-teen person and all ten expressions directly into one TAR:

python tools/infer_character_tar.py --person person-0253 \
  --plan metadata/person-plan.json --output person-0253-emotions.tar

--base-path can point to an already downloaded base checkpoint. --no-compile disables compilation when needed. The default inference helper fuses the LoRA and QKV projections, compiles repeated transformer blocks, batches the positive/negative CFG branches and caches text embeddings. These optimizations preserve 50-step real CFG4 sampling; there is no denoising approximation cache. The pipeline is BF16 and expects a CUDA GPU with sufficient memory. Published seeds and revisions enable matching the generation recipe, but bitwise reproduction across GPU/runtime versions is not promised.

Timing, validation and provenance

Production uses eight independent A100-SXM4-80GB workers, batch size one per GPU. Neutral generation and reference edits have different workloads; their measured means are reported separately in performance. Reference edits include reference VAE encoding in the timed call. Positive prompt preencoding, initial compilation, loading, PNG encoding, TAR writes and uploads are excluded from those inference timings. The pilot shared one GPU with background generation and used CPU offload, so its timings are excluded from dedicated-worker statistics.

Every closed shard is verified against remote SHA256 and byte size after upload. The final integrity report fully decodes all PNGs, checks adjacent PNG/JSON/TXT grouping, unique planned IDs, prompt/seed equality, image checksums and reference linkage, complete ten-emotion person groups, and the exact planned distributions. Integrity report and visual review are published when their respective checks are complete.

The character adapter was trained on the previously selected Vivarium greenscreen data and chosen at training step 1,500, rank/alpha 64, strength 1.0. This release's 500 people are newly planned synthetic identities. Their count distribution is independent of the original Anim-E procedural prompt row frequencies. Generation terms are documented in LICENSE.md with the full model license.