Image generation (Skill V3)

SkillDocs & knowledge

This skill is not “generate a pretty picture”. It is a plate that executes the brief. Overlay type stays on the board (createtext). Kernel still Decide → Paint → Observe → Review. This skill owns how the bitmap is specified.

Then ask your AI: use the Image generation (Skill V3)

Details

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Add it as a step in a pipeline you build on Ahel, or connect Ahel to Claude, ChatGPT, Cursor, Claude Code or Codex and ask your AI to use it.

About this skill

zuoge, open-source AI design workspace with a document-derived SoA infinite artboard and Design Agent

What this skill tells your AI

The instructions your AI receives, as published by recombyn/zuoge in skills/domains/image_gen/SKILL.md and read by Ahel’s review.

This skill is not “generate a pretty picture”. It is a plate that executes the brief. Overlay type stays on the board (create_text). Kernel still Decide → Paint → Observe → Review. This skill owns how the bitmap is specified.

Helper: may load with a surface (poster_craft / landing_page / …). It does not replace them.

Process (mandatory order)

DESIGN BRIEF
 → VISUAL THESIS
 → COMPOSITION
 → MATERIAL
 → LIGHTING
 → IMAGE

Do not jump to a prompt until thesis, crop, material, and light are named.

1. Brief → thesis

Reuse P0 design_brief. The image must argue visual_thesis and show visual_hero. Style DNA (material / lighting / imagery) is the prompt spine — not “cinematic, 8k, masterpiece”.

2. Composition (in the plate)

Pick crop and focal before generating:

  • What is in frame vs left to overlay type / vector chrome
  • One primary subject; quiet bands where titles will sit
  • Camera: distance, angle, what is cut off

If the board composition is center_hero, the plate should not also scream a second hero in the corner.

3. Material → lighting → image

Name both in the prompt (from Style DNA):

SlotAsk
Materialjade / aged steel / paper / cloth / skin / glass / …
Lightingdirection, hardness, color of light, what it reveals
Mediumphoto / illustration / grain print / quiet 3D — one

Then emit create_image (hydrate). Simple geometry that stays crisp may stay vector instead — crude shape piles are not atmosphere.

4. Forbidden: baked title

Do not render titles, slogans, watermarks, or UI chrome inside the bitmap unless the user explicitly asks for text in the picture (lettering-as-image, packaging, neon sign, 书法字形).

Default: overlay catalog type on a quiet band. Lettering plates are the exception, not the habit.

5. Honesty

Unless the user asks, do not invent logos, prices, phones, or readable fake copy in the pixels.

6. Cutout vs full-bleed

  • Full-bleed atmosphere: keep the plate; type sits on quiet zones.
  • Product / lettering on a colored board: cutout so no leftover white box.

Hard rules

  1. Brief + thesis before generate.
  2. Composition + material + lighting in the prompt — not style-spam adjectives.
  3. No baked titles unless explicitly requested.
  4. Anti-slop: unjustified purple-blue glass postcard heroes fail Review.

Done when

  • Plate matches thesis at a glance (subject + material + light)
  • Room for overlay type (or user-requested in-image lettering only)
  • No invented on-image copy

Related

Surfaces via Decide (poster_craft, landing_page, banner_ad, …) · icon_set for many marks · foundations via extends

Signals

GitHub stars
55
Forks
9
Last commit
Sep 2026
Advanced
Item type
skill
Key
image-gen-recombyn
Source
github.com/recombyn/zuoge