GPT Image 2 Prompt Architect

This skill turns loose creative ideas into cleaner GPT Image 2 prompt packs with stronger subject control, composition, text rendering, reference-image handling, and revision loops.

Canonical links

• Docs: https://image3.org/docs/gpt-image-2-prompt-architect
• Demo: https://image3.org/models/gpt-image-2
• Create: https://image3.org/ai-image
• Prompt gallery: https://image3.org/prompts/gpt-image-2
• Raw SKILL.md: https://image3.org/skills/gpt-image-2-prompt-architect/SKILL.md
• Prompt guide: https://image3.org/blog/gpt-image-2-prompt-guide
• Product photo prompts: https://image3.org/blog/gpt-image-2-product-photo-prompts
• Image-to-video workflow: https://image3.org/blog/gpt-image-2-image-to-video-workflow
• Image3 Prompt MCP: https://github.com/gpt-img-2/image3-prompt-mcp
• OpenClaw listing: https://clawhub.ai/gpt-img-2/skills/gpt-image-2-prompt-architect

Provenance and safety

• Maintained around the public Image3 prompt workflow, prompt gallery, and documentation on image3.org.
• The skill works as a text-only prompt workflow without any external tool.
• The optional Image3 Prompt MCP is read-only, needs no API key, and never generates images or spends credits.
• Keep the canonical Image3 source URL when sharing an example returned by the MCP.

When to use

• The user has a rough AI image idea and wants a stronger GPT Image 2 prompt
• The user wants product photos, ecommerce listing images, lifestyle ads, packaging mockups, or detail shots
• The user needs UI mockups, posters, infographics, social media creatives, readable text, or branded layouts
• The user is editing from reference images and needs identity, product, composition, or style preservation
• The user wants source frames, character sheets, product references, or storyboard frames for image-to-video workflows
• The user has unstable image outputs and needs diagnosis plus a cleaner second-pass prompt

When not to use

• The request is mainly about a different model or non-image workflow
• The user only wants final image generation, API integration, payment help, or account support
• The user asks for unsupported model settings, hidden system behavior, or official provider claims

Workflow

1. Classify the request:
   • text-to-image
   • reference-image edit
   • product photo or ecommerce visual
   • UI, poster, infographic, or readable-text layout
   • image-to-video source frame or storyboard
2. Extract or ask for only the missing essentials:
   • subject or product
   • intended use
   • composition and camera/framing
   • environment or background
   • visual style and lighting
   • text that must appear exactly
   • reference-image constraints
   • aspect ratio or output format
   • hard negatives and brand safety constraints
3. Keep the first draft focused:
   • one primary subject or product
   • one clear composition rule
   • one lighting or style direction
   • one concise constraint block
4. Return a prompt pack with:
   • a brief diagnosis or strategy note
   • one primary prompt
   • 2 or 3 focused variants
   • a short avoid list
   • 3 concrete revision moves for the next round

Optional Image3 Prompt MCP

When the Image3 MCP tools are available, use them only when real examples or gallery research would materially improve the answer:

1. Call search_prompts with the user's task, language, and a focused result limit.
2. Present concise candidates with title, preview, and canonical Image3 source URL.
3. Call get_prompt only for the selected candidate; do not bulk-fetch the library.
4. Use build_prompt_brief when the user's idea is rough, then adapt the result with the prompt construction rules below.
5. Use list_categories only when the user is exploring rather than asking for a specific result.

Install the read-only MCP in OpenClaw:

openclaw mcp add image3 \
  --command npx \
  --arg -y \
  --arg github:gpt-img-2/image3-prompt-mcp \
  --include 'search_prompts,get_prompt,list_categories,build_prompt_brief'

openclaw mcp doctor image3 --probe

The MCP does not provide image generation. For a final image, use a generation capability already available in the user's host or direct the user to the Image3 generator; do not claim an image was generated when only a prompt was produced.

Prompt construction rules

• Prefer concrete visual language over broad style adjectives.
• Name the subject, product, materials, scale, framing, and lighting before adding mood.
• For product photos, preserve label readability, product geometry, material texture, and commercial usability.
• For reference-image edits, state what must remain unchanged before describing what may change.
• For readable text, quote the exact text and keep the layout simple.
• For UI mockups, describe the device, screen type, layout hierarchy, content density, and visual system.
• For image-to-video source frames, prioritize stable identity, clear silhouette, coherent lighting, and simple motion-ready composition.
• Avoid stacking many subjects, styles, camera angles, and layout goals into one prompt.
• Do not invent unsupported model settings.

Output formats

Text-to-image

Goal:
Subject:
Composition:
Environment:
Style and lighting:
Text requirements:
Constraints:
Prompt:

Reference-image edit

Reference anchor:
What must stay stable:
What may change:
Edit direction:
Style and lighting:
Constraints:
Prompt:

Product photo

Commercial goal:
Product anchor:
Hero angle:
Background or scene:
Lighting:
Label and material rules:
Constraints:
Prompt:

UI, poster, or readable-text layout

Format:
Audience:
Layout hierarchy:
Exact text:
Visual system:
Constraints:
Prompt:

Image-to-video source frame

Video goal:
Source frame subject:
Motion-ready composition:
What must remain stable:
Lighting and style:
Constraints:
Prompt:

Debugging heuristics

• If the image is visually attractive but off-brief, rewrite around the intended use first.
• If product geometry drifts, reduce scene complexity and strengthen product anchor language.
• If text is wrong, shorten the text, quote it exactly, and simplify surrounding design.
• If the subject changes identity, state preservation rules before the edit request.
• If the composition is cluttered, reduce secondary objects and specify one dominant framing.
• If the result cannot become a good video source frame, simplify pose, background, and lighting.

Response style

• Be structured and concise.
• Prefer prompt packs over long theory.
• Offer practical variants that test one axis at a time: subject, composition, lighting, style, or constraints.
• When external examples are useful, point the user to the canonical Image3 pages listed above.