generate-prompt
Generate optimized image generation prompts for a specific interior design style and AI provider. Use when the user wants to create prompts for rendering interior designs.
SKILL.md.claude/skills/generate-prompt/SKILL.md
GitHubGenerate Interior Design Prompt
Create optimized image generation prompts tailored to specific AI providers and design styles.
Preprocessing Check (MANDATORY — run before generating prompts)
Before generating any prompt, detect whether this project has preprocessed control maps. This determines prompt strategy.
!ls projects/${PROJECT_NAME}/references/preprocessed/ 2>/dev/null && echo "HAS_CONTROL_MAPS=true" || echo "HAS_CONTROL_MAPS=false"
!ls projects/${PROJECT_NAME}/references/masks/ 2>/dev/null && echo "HAS_MASKS=true" || echo "HAS_MASKS=false"
!ls projects/${PROJECT_NAME}/references/*.{jpg,jpeg,png,webp} 2>/dev/null && echo "HAS_REFERENCE=true" || echo "HAS_REFERENCE=false"
Decision Logic
HAS_REFERENCE=true AND HAS_CONTROL_MAPS=false?
└── STOP. Run /preprocess-room FIRST. Do NOT generate prompts without control maps when a reference photo exists.
HAS_CONTROL_MAPS=true?
└── STRUCTURE CONTROL MODE:
- Prompt must include spatial/structural keywords: "preserving room layout", "maintaining spatial structure"
- Prompt must emphasize materials, colors, furniture OVER room geometry (the depth map handles geometry)
- Do NOT describe wall positions, window placement, ceiling height — the control map provides this
- For Stability AI: prompt focuses on style + materials (structure API handles layout)
- For Gemini editing: pass depth map as reference image alongside the style prompt
HAS_MASKS=true?
└── INPAINTING MODE:
- Generate TWO prompts: one for the masked region only, one for full scene context
- Masked region prompt: describe ONLY what goes in the masked area (e.g., "a tufted velvet sofa in navy blue")
- Full context prompt: describe the whole room for coherence, but mark what changes
Neither (text-to-image)?
└── Standard mode — describe everything including room geometry, layout, spatial arrangement
Mode selection is automatic. Adjust prompt strategy silently based on what exists.
Input Requirements
Gather from the user (ask if not provided via $ARGUMENTS):
- Room type: living room, bedroom, kitchen, bathroom, office, dining room, etc.
- Design style: reference
/style-guidefor available styles - Target provider: Gemini, OpenAI/DALL-E, Grok, Stability AI, Midjourney, or "all"
- Additional details (optional): room dimensions, specific furniture, color preferences, lighting mood, camera angle
Prompt Structure Per Provider
OpenAI (DALL-E 3 / GPT-4o)
[Room type] interior design in [style] style. [Detailed description of furniture, materials, colors].
Lighting: [lighting type]. Camera angle: [angle]. Photorealistic, 8K, interior photography, architectural digest quality.
- Keep under 400 characters for DALL-E 3
- Be explicit about photorealism
- Specify camera angle for composition
Google Gemini (Imagen)
A photorealistic interior photograph of a [room type] designed in [style] style.
Features: [materials], [furniture], [colors], [textures].
Atmosphere: [lighting], [mood].
Style: professional interior photography, high resolution, detailed textures.
- Gemini responds well to structured, descriptive prompts
- Emphasize "photograph" for realism
Stability AI (Stable Diffusion)
Positive: photorealistic interior design, [room type], [style] style, [materials], [colors], [furniture], [lighting], professional photography, 8k uhd, ray tracing, detailed textures
Negative: cartoon, drawing, sketch, low quality, blurry, distorted, watermark, text
- Always include negative prompt
- Use comma-separated keyword style
- Include quality boosters: 8k, ray tracing, detailed
Midjourney
[Room type] interior, [style] design style, [key materials and colors], [lighting mood], professional interior photography --ar 16:9 --v 6 --style raw --q 2
- Use
--arfor aspect ratio (16:9 for wide rooms, 4:3 for detail shots) --style rawfor photorealism- Keep descriptive, flowing language
Grok (xAI)
Generate a photorealistic image of a [room type] with [style] interior design.
Include: [specific furniture], [materials like marble/wood/fabric], [color palette].
Mood: [warm/cool/dramatic/serene]. Lighting: [natural/artificial/mixed].
Output Format
Present the generated prompts in a clear format:
- Show the prompt for each requested provider
- Highlight style-specific keywords used (from style guide)
- Suggest 2-3 variations (different camera angles, lighting moods, or detail focus)
- Note any provider-specific limitations or tips
Session Relevant Skills
/render— the natural next step. Takes these prompts and sends them to AI providers. Suggest specific providers based on the style (e.g., Stability for texture-heavy styles, DALL-E for clean modern looks)./style-guide— essential reference during prompt generation. Always pull keywords, materials, and colors from the style guide to ensure accuracy. Don't rely on memory — styles have specific vocabulary./mood-board— if the user came here without a mood board, the prompts may lack atmospheric depth. Suggest /mood-board first for complex or ambiance-heavy styles (Wabi-Sabi, Bohemian, Biophilic)./design-consult— if requirements are unclear (no room type, no style, no preferences), redirect to /design-consult rather than guessing./compare-models— after /render with multiple providers, /compare-models evaluates which output is best./edit-design— if the user has a reference image and wants prompts based on it, redirect to /edit-design which handles scene analysis before prompt generation.
Gotchas
- Provider prompt lengths differ wildly: DALL-E 3 = under 400 chars. Midjourney = flowing description. Stability = comma-separated keywords. Don't use one format for all providers.
- Negative prompts are not universal: Only Stability AI uses explicit negative prompts. Don't add negative prompt sections for DALL-E or Gemini — they ignore or misinterpret them.
- "All providers" is expensive: Generating for all providers means multiple API calls. Ask the user if they really want all, or suggest 2-3 best providers for their style.
- Style keywords order matters: Put the most important style descriptors first. AI models weight earlier tokens more heavily. Room type and style should lead, decorative details follow.