README.md

---
language:
- en
license: other
license_name: flux-output-conditions
license_link: LICENSE.md
task_categories:
- text-to-image
tags:
- synthetic
- characters
- greenscreen
- webdataset
- flux
- animation
size_categories:
- 1K<n<10K
pretty_name: Vivarium Greenscreen Characters — FLUX 9B
configs:
- config_name: default
  data_files:
  - split: train
    path:
    - job-data/*.tar
    - data/pilot-person-0002.tar
    - data/pilot-person-0120.tar
    - data/pilot-person-0253.tar
    - data/train-g0-00000.tar
    - data/train-g0-00001.tar
    - data/train-g0-00002.tar
    - data/train-g0-00003.tar
    - data/train-g0-00004.tar
    - data/train-g0-00005.tar
    - data/train-g0-00006.tar
    - data/train-g0-00007.tar
    - data/train-g0-00009.tar
    - data/train-g0-00011.tar
    - data/train-g0-00012.tar
    - data/train-g0-00013.tar
    - data/train-g0-00014.tar
    - data/train-g0-00015.tar
    - data/train-g0-00016.tar
    - data/train-g0-00017.tar
    - data/train-g0-00018.tar
    - data/train-g0-00019.tar
    - data/train-g0-00020.tar
    - data/train-g0-00021.tar
    - data/train-g0-00022.tar
    - data/train-g0-00023.tar
    - data/train-g0-00024.tar
    - data/train-g0-00025.tar
    - data/train-g0-00026.tar
    - data/train-g0-00027.tar
    - data/train-g0-00028.tar
    - data/train-g0-00029.tar
    - data/train-g0-00030.tar
    - data/train-g0-00031.tar
    - data/train-g1-00000.tar
    - data/train-g1-00001.tar
    - data/train-g1-00002.tar
    - data/train-g1-00003.tar
    - data/train-g1-00004.tar
    - data/train-g1-00005.tar
    - data/train-g1-00006.tar
    - data/train-g1-00007.tar
    - data/train-g1-00008.tar
    - data/train-g1-00009.tar
    - data/train-g1-00010.tar
    - data/train-g1-00011.tar
    - data/train-g1-00012.tar
    - data/train-g1-00013.tar
    - data/train-g1-00014.tar
    - data/train-g1-00015.tar
    - data/train-g1-00016.tar
    - data/train-g1-00017.tar
    - data/train-g1-00019.tar
    - data/train-g1-00020.tar
    - data/train-g1-00021.tar
    - data/train-g1-00022.tar
    - data/train-g1-00023.tar
    - data/train-g1-00024.tar
    - data/train-g1-00025.tar
    - data/train-g1-00026.tar
    - data/train-g1-00027.tar
    - data/train-g1-00028.tar
    - data/train-g1-00029.tar
    - data/train-g1-00030.tar
    - data/train-g1-00031.tar
    - data/train-g2-00000.tar
    - data/train-g2-00001.tar
    - data/train-g2-00003.tar
    - data/train-g2-00004.tar
    - data/train-g2-00005.tar
    - data/train-g2-00006.tar
    - data/train-g2-00007.tar
    - data/train-g2-00008.tar
    - data/train-g2-00009.tar
    - data/train-g2-00010.tar
    - data/train-g2-00011.tar
    - data/train-g2-00012.tar
    - data/train-g2-00013.tar
    - data/train-g2-00014.tar
    - data/train-g2-00015.tar
    - data/train-g2-00016.tar
    - data/train-g2-00017.tar
    - data/train-g2-00018.tar
    - data/train-g2-00019.tar
    - data/train-g2-00020.tar
    - data/train-g2-00021.tar
    - data/train-g2-00022.tar
    - data/train-g2-00023.tar
    - data/train-g2-00024.tar
    - data/train-g2-00025.tar
    - data/train-g2-00026.tar
    - data/train-g2-00027.tar
    - data/train-g2-00028.tar
    - data/train-g2-00029.tar
    - data/train-g2-00030.tar
    - data/train-g2-00031.tar
    - data/train-g2-00032.tar
    - data/train-g3-00000.tar
    - data/train-g3-00001.tar
    - data/train-g3-00002.tar
    - data/train-g3-00003.tar
    - data/train-g3-00004.tar
    - data/train-g3-00006.tar
    - data/train-g3-00007.tar
    - data/train-g3-00008.tar
    - data/train-g3-00009.tar
    - data/train-g3-00010.tar
    - data/train-g3-00011.tar
    - data/train-g3-00012.tar
    - data/train-g3-00013.tar
    - data/train-g3-00014.tar
    - data/train-g3-00015.tar
    - data/train-g3-00016.tar
    - data/train-g3-00017.tar
    - data/train-g3-00018.tar
    - data/train-g3-00019.tar
    - data/train-g3-00020.tar
    - data/train-g3-00021.tar
    - data/train-g3-00022.tar
    - data/train-g3-00023.tar
    - data/train-g3-00024.tar
    - data/train-g3-00025.tar
    - data/train-g3-00026.tar
    - data/train-g3-00027.tar
    - data/train-g3-00028.tar
    - data/train-g3-00029.tar
    - data/train-g3-00030.tar
    - data/train-g3-00031.tar
    - data/train-g4-00000.tar
    - data/train-g4-00001.tar
    - data/train-g4-00002.tar
    - data/train-g4-00003.tar
    - data/train-g4-00004.tar
    - data/train-g4-00005.tar
    - data/train-g4-00006.tar
    - data/train-g4-00007.tar
    - data/train-g4-00008.tar
    - data/train-g4-00009.tar
    - data/train-g4-00010.tar
    - data/train-g4-00011.tar
    - data/train-g4-00012.tar
    - data/train-g4-00013.tar
    - data/train-g4-00014.tar
    - data/train-g4-00015.tar
    - data/train-g4-00016.tar
    - data/train-g4-00017.tar
    - data/train-g4-00018.tar
    - data/train-g4-00019.tar
    - data/train-g4-00020.tar
    - data/train-g4-00021.tar
    - data/train-g4-00022.tar
    - data/train-g4-00023.tar
    - data/train-g4-00024.tar
    - data/train-g4-00025.tar
    - data/train-g4-00026.tar
    - data/train-g4-00027.tar
    - data/train-g4-00028.tar
    - data/train-g4-00029.tar
    - data/train-g4-00030.tar
    - data/train-g4-00031.tar
    - data/train-g5-00000.tar
    - data/train-g5-00001.tar
    - data/train-g5-00002.tar
    - data/train-g5-00003.tar
    - data/train-g5-00004.tar
    - data/train-g5-00006.tar
    - data/train-g5-00007.tar
    - data/train-g5-00008.tar
    - data/train-g5-00009.tar
    - data/train-g5-00010.tar
    - data/train-g5-00011.tar
    - data/train-g5-00012.tar
    - data/train-g5-00013.tar
    - data/train-g5-00014.tar
    - data/train-g5-00015.tar
    - data/train-g5-00016.tar
    - data/train-g5-00017.tar
    - data/train-g5-00018.tar
    - data/train-g5-00019.tar
    - data/train-g5-00020.tar
    - data/train-g5-00021.tar
    - data/train-g5-00022.tar
    - data/train-g5-00023.tar
    - data/train-g5-00024.tar
    - data/train-g5-00025.tar
    - data/train-g5-00026.tar
    - data/train-g5-00027.tar
    - data/train-g5-00028.tar
    - data/train-g5-00029.tar
    - data/train-g5-00030.tar
    - data/train-g6-00000.tar
    - data/train-g6-00001.tar
    - data/train-g6-00002.tar
    - data/train-g6-00003.tar
    - data/train-g6-00004.tar
    - data/train-g6-00005.tar
    - data/train-g6-00006.tar
    - data/train-g6-00007.tar
    - data/train-g6-00008.tar
    - data/train-g6-00009.tar
    - data/train-g6-00010.tar
    - data/train-g6-00011.tar
    - data/train-g6-00012.tar
    - data/train-g6-00013.tar
    - data/train-g6-00014.tar
    - data/train-g6-00015.tar
    - data/train-g6-00016.tar
    - data/train-g6-00017.tar
    - data/train-g6-00018.tar
    - data/train-g6-00019.tar
    - data/train-g6-00020.tar
    - data/train-g6-00021.tar
    - data/train-g6-00022.tar
    - data/train-g6-00023.tar
    - data/train-g6-00024.tar
    - data/train-g6-00025.tar
    - data/train-g6-00026.tar
    - data/train-g6-00027.tar
    - data/train-g6-00028.tar
    - data/train-g6-00029.tar
    - data/train-g6-00030.tar
    - data/train-g6-00031.tar
    - data/train-g7-00000.tar
    - data/train-g7-00001.tar
    - data/train-g7-00002.tar
    - data/train-g7-00003.tar
    - data/train-g7-00004.tar
    - data/train-g7-00005.tar
    - data/train-g7-00006.tar
    - data/train-g7-00007.tar
    - data/train-g7-00008.tar
    - data/train-g7-00009.tar
    - data/train-g7-00010.tar
    - data/train-g7-00011.tar
    - data/train-g7-00012.tar
    - data/train-g7-00013.tar
    - data/train-g7-00014.tar
    - data/train-g7-00015.tar
    - data/train-g7-00016.tar
    - data/train-g7-00017.tar
    - data/train-g7-00018.tar
    - data/train-g7-00019.tar
    - data/train-g7-00020.tar
    - data/train-g7-00021.tar
    - data/train-g7-00022.tar
    - data/train-g7-00023.tar
    - data/train-g7-00024.tar
    - data/train-g7-00025.tar
    - data/train-g7-00026.tar
    - data/train-g7-00027.tar
    - data/train-g7-00028.tar
- config_name: jobs
  data_files:
  - split: train
    path: job-data/*.tar
dataset_info:
- config_name: default
  features: &id001
  - name: png
    dtype: image
  - name: json
    struct:
    - name: personality_id
      dtype: string
    - name: character_id
      dtype: string
    - name: age_group
      dtype: string
    - name: age_years
      dtype: int64
    - name: age_months
      dtype: int64
    - name: gender
      dtype: string
    - name: ethnicity
      dtype: string
    - name: ethnicity_prompt_label
      dtype: string
    - name: skin_tone
      dtype: string
    - name: hair
      dtype: string
    - name: build
      dtype: string
    - name: outfit_id
      dtype: string
    - name: outfit_style
      dtype: string
    - name: outfit
      dtype: string
    - name: seed
      dtype: int64
    - name: default_emotion
      dtype: string
    - name: description
      dtype: string
    - name: safety_regeneration_attempt
      dtype: int64
    - name: safety_regeneration_policy_sha256
      dtype: string
    - name: seed_strategy
      dtype: string
    - name: safety_reference_strategy
      dtype: string
    - name: id
      dtype: string
    - name: emotion
      dtype: string
    - name: prompt
      dtype: string
    - name: negative_prompt
      dtype: string
    - name: reference_id
      dtype: string
    - name: schema_version
      dtype: string
    - name: generated_utc
      dtype: string
    - name: image_sha256
      dtype: string
    - name: reference_image_sha256
      dtype: string
    - name: reference_resize
      struct:
      - name: width
        dtype: int64
      - name: height
        dtype: int64
      - name: resample
        dtype: string
    - name: inference
      struct:
      - name: model
        dtype: string
      - name: revision
        dtype: string
      - name: lora
        dtype: string
      - name: lora_revision
        dtype: string
      - name: lora_subfolder
        dtype: string
      - name: lora_sha256
        dtype: string
      - name: lora_rank
        dtype: int64
      - name: lora_alpha
        dtype: int64
      - name: lora_training_step
        dtype: int64
      - name: lora_scale
        dtype: float64
      - name: steps
        dtype: int64
      - name: guidance_scale
        dtype: float64
      - name: seed
        dtype: int64
      - name: width
        dtype: int64
      - name: height
        dtype: int64
      - name: dtype
        dtype: string
      - name: quantization
        dtype: bool
      - name: distillation
        dtype: bool
      - name: max_sequence_length
        dtype: int64
      - name: cfg_mode
        dtype: string
      - name: lora_fused
        dtype: bool
      - name: qkv_fused
        dtype: bool
      - name: compile_mode
        dtype: string
      - name: negative_embedding_cached
        dtype: bool
      - name: positive_embedding_cached
        dtype: bool
    - name: runtime
      struct:
      - name: profile
        dtype: string
      - name: gpu
        dtype: string
      - name: physical_gpu
        dtype: int64
      - name: batch_size
        dtype: int64
      - name: generation_seconds
        dtype: float64
      - name: positive_text_encoding_seconds
        dtype: float64
      - name: torch
        dtype: string
      - name: cuda
        dtype: string
      - name: includes_compile_warmup
        dtype: bool
      - name: gpu_shared_with_backgrounds
        dtype: bool
      - name: model_cpu_offload
        dtype: bool
      - name: gpu_shared_with_character_production
        dtype: bool
    - name: storage
      struct:
      - name: shard
        dtype: string
      - name: image_member
        dtype: string
      - name: json_member
        dtype: string
      - name: prompt_member
        dtype: string
      - name: encoding
        dtype: string
    - name: safety_check
      struct:
      - name: max_score
        dtype: float64
      - name: attempt
        dtype: int64
      - name: recipe_variant
        dtype: int64
      - name: utc
        dtype: string
      - name: id
        dtype: string
      - name: passed
        dtype: bool
      - name: image_sha256
        dtype: string
      - name: generation_seconds
        dtype: float64
      - name: screening_seconds
        dtype: float64
      - name: positive_text_encoding_seconds
        dtype: float64
      - name: retained_original
        dtype: bool
      - name: views
        struct:
        - name: original
          struct:
          - name: clip_nsfw
            dtype: float64
          - name: efficientnet_nsfw
            dtype: float64
          - name: efficientnet_classes
            struct:
            - name: drawing
              dtype: float64
            - name: hentai
              dtype: float64
            - name: neutral
              dtype: float64
            - name: porn
              dtype: float64
            - name: sexy
              dtype: float64
        - name: foreground
          struct:
          - name: clip_nsfw
            dtype: float64
          - name: efficientnet_nsfw
            dtype: float64
          - name: efficientnet_classes
            struct:
            - name: drawing
              dtype: float64
            - name: hentai
              dtype: float64
            - name: neutral
              dtype: float64
            - name: porn
              dtype: float64
            - name: sexy
              dtype: float64
    - name: record_type
      dtype: string
    - name: production_phase
      dtype: string
    - name: lora_scale
      dtype: float64
    - name: special_class
      dtype: string
    - name: special_variant
      dtype: string
    - name: composition
      dtype: string
    - name: job_id
      dtype: string
    - name: job_key
      dtype: string
    - name: profession_family
      dtype: string
    - name: job_title_de
      dtype: string
    - name: job_title_en
      dtype: string
    - name: job_outfit
      dtype: string
    - name: job_outfit_index
      dtype: int64
    - name: base_outfit_id
      dtype: string
    - name: base_outfit_style
      dtype: string
    - name: base_outfit
      dtype: string
    - name: base_description
      dtype: string
    - name: reference_source_phase
      dtype: string
    - name: reference_source_repo_path
      dtype: string
    - name: reference_source_shard
      dtype: string
    - name: reference_source_member
      dtype: string
    - name: reference_source_generated_utc
      dtype: string
  - name: txt
    dtype: string
  - name: __key__
    dtype: string
  - name: __url__
    dtype: string
- config_name: jobs
  features: *id001
---

# Complete character collection and professional outfit plan

**Planned total: 840 identities, 12,140 images.** The expression/special collection contains 7,500 planned images. A further **464 adult humans × 10 professional outfits = 4,640 sprites** reuse those identities and their neutral references; they do not add new personality IDs. See [all 100 jobs and the exact outfit allocation](JOB_OUTFITS.md). Job sprites uploaded so far: **16/4,640**. The `jobs` streaming configuration becomes available when its first TAR is uploaded.

The [teen/child detector recheck](SAFETY_RECHECK.md) reports actual image checksums, both LAION classifier scores, operational thresholds and any replacements. Intended ages, professions and demographic attributes describe prompting, not inferred real-world classifications.


# Vivarium warm-painted greenscreen characters

**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.

**Production snapshot: 500/500 complete people; 4860/5,000 images uploaded.** Pilot and production validation are recorded in the linked metadata. Counts describe the snapshot; incremental shard commits can be newer.

The [base model](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-9B) and [character LoRA with examples and instructions](https://huggingface.co/laion/FLUX.2-klein-base-9B-vivarium-loras) are linked separately. The compatible [background dataset](https://huggingface.co/datasets/laion/Anim-E-Ghibli-Backgrounds-FLUX-9B) 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](CLASSES.md), [taxonomy](metadata/taxonomy.json), [exact allocation](metadata/allocation.json), [500-person CSV](metadata/persons.csv), and [full prompt plan](metadata/person-plan.json) 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](metadata/outfit-allocation-refinement.json) 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](SAFETY_RECHECK.md).

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](metadata/skin-tone-pilot.json) and [prompt refinement audit](metadata/skin-tone-prompt-refinement.json) 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:

```text
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:

```text
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:

```text
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](metadata/shards.json), [inference records](metadata/inference-records.jsonl), [progress](metadata/progress.json), and [sample index](metadata/sample-index.json) 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:

```python
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 card config selects only `data/*.tar` so index files are not mistaken for training examples. An alternate standard-library TAR/HTTP reader avoids extracting image files:

```bash
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](metadata/provenance.json); 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:

```bash
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](metadata/performance.json). 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](metadata/integrity-validation.json) and [visual review](metadata/quality-review.json) 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](LICENSE.md) with the full [model license](MODEL_LICENSE.md).