Consistent Asset Sets: one approved reference, style files and review sheets for AI-made art, props and cards

Match the attached reference exactly in style, framing and light. Change only: [delta from data row].
No text in the image.

Deltas drift less than fresh descriptions.

Write a script that renders every item in [folder] at the same size and camera, tiles them in a grid with reference/REF-001 top-left and names underneath.
Save it to review/sheet-[date].png.

Drift hides one at a time and shows on a sheet.

# Consistent Asset Sets: one approved reference, style files and review sheets for AI-made art, props and cards

A PaperChaseWebb pack, version 1.0. Written for Claude, ChatGPT, Claude Code and Codex, and plain Markdown, so the style file, the reference lock method, the review sheet and the prompts work with any image, 3D or coding tool you already use.

Making one good image with AI is easy now. Making forty that look like they belong together is not. Card sets drift in line weight. Props for one game come out at five different scales. A client's campaign images slowly change palette from Monday to Friday. The fix is not a better prompt. It is a small production system: approve one piece, lock it as the reference, write the style down, make the rest against it, and review the whole set on one sheet.

This pack is that system, for three ways of making assets: generating them, writing them as code and picking from premade libraries. It is built from public build reports by people who made card sets, game props, 3D previs and full game art sets with agents, plus our own client asset work. Sources are credited by link in the Sources section.

## What is inside

| Section | What it is for |
|---|---|
| 1. Why sets drift | Five causes, and which step in this pack fixes each |
| 2. Choose a mode per asset | Generate, code or premade, with a tradeoff table |
| 3. The reference lock method | Perfect one, lock it, attach it to every request, ask for deltas only |
| 4. STYLE.md | A style file template with palette, line, light, camera and forbidden list |
| 5. PROMPTS.json | A saved prompt library so good prompts are reused, not retyped |
| 6. Assets as code | Script-built props, asset description files and the annotate loop |
| 7. The review sheet | One image of the whole set, plus a ten-point check |
| 8. 3D as a camera guide | Use a rough 3D model to steer generated video and stills |
| 9. Cost and rights ledgers | Track spend per stage and the license on every asset |
| 10. Client brand sets | Turn a brand guide into STYLE.md and an approval flow |
| 11. Rules block | A paragraph for CLAUDE.md, AGENTS.md or a project |
| 12. The set-reviewer skill | A SKILL.md that checks a batch against the reference |
| 13. Prompts | 12 prompts |
| 14. Sources | What we used and how we used it |