adv-art

ADV Art Pipeline

Turn story requirements into reproducible, licensed, runtime-ready visual assets. Keep logical asset identity separate from versioned public URLs.

Workflow

1. Freeze an art bible

Read character cards, scene cards, the story outline, and existing UI proportions. Define:

• shared style, line quality, palette, lighting, period, camera language, and prohibited motifs;
• character silhouette, body proportions, clothing, fixed accessories, color swatches, and pose rules;
• a required expression matrix per visible character;
• background aspect ratios and which moments require one-off CG rather than reusable scenery.

Do not generate final images until the recurring character design and expression names are stable.

2. Plan logical assets

Use stable lowercase IDs. Keep pose and expression names semantic (standing/sad), not sequential (image-07). Add every required character expression to the manifest before generation so missing variants are detectable.

Read references/asset-manifest.md before creating the manifest.

3. Generate and review originals

When an image-generation capability is available, generate original assets from the art bible and each card's imagePrompt. Keep character identity, outfit, crop, lighting, and transparent-background requirements explicit in every expression request.

Inspect each result before accepting it. Reject identity drift, accidental text, extra limbs, clipped ears/hands/tails, inconsistent costume details, opaque halos, or backgrounds embedded in tachie. Preserve the model/tool name, prompt version, generation date, and content license in provenance metadata.

For scene backgrounds in a local Codex workflow, read references/codex-authoring.md and use the task/candidate protocol. Codex built-in image generation is the preferred first executor when available, but it is not part of the asset domain contract. A successful generation is only a candidate; never register it before the user visually confirms that exact file.