| name | prompt-craft |
| description | Translate natural language image descriptions into detailed, structured DALL-E prompts with subject, style, composition, lighting, and mood specifications.
|
Image prompt crafting skill.
When to activate
Use this skill when you need to construct a prompt for generate_image
from a user's description. Always activate this before calling
generate_image.
Methodology
1. Extract intent
Parse the user's description for:
- Subject: what is in the image (person, object, scene, abstract)
- Style: art style (photorealistic, illustration, watercolor, etc.)
- Mood: emotional tone (calm, energetic, mysterious, professional)
- Colors: specific colors or palette preferences
- Composition: framing, perspective, layout
- Context: where will this image be used (logo, social media,
presentation, personal)
If any of these are missing and matter for the result, ask the user.
2. Check memory
Recall style preferences and brand guidelines:
recall("style_preference")
recall("brand_guideline")
Apply any saved preferences unless the user's current request
explicitly overrides them.
3. Build the prompt
Construct the prompt using this structure:
[Subject description], [style], [composition], [lighting], [mood], [details]
Rules for effective prompts:
- Be specific about the subject: "a golden retriever puppy sitting
on a red cushion" not "a dog"
- Name the art style explicitly: "digital illustration", "oil painting
style", "35mm film photography", "vector art"
- Specify lighting: "soft natural light", "dramatic side lighting",
"neon glow", "golden hour"
- Include composition cues: "close-up", "wide angle", "bird's eye
view", "centered", "rule of thirds"
- Add atmosphere: "moody", "minimalist", "vibrant", "muted tones"
- Keep prompts under 400 characters for best results
- Avoid negations ("no background") -- describe what you want, not
what you do not want
4. Select parameters
Choose the right parameters for the use case:
| Use case | Size | Style | Quality |
|---|
| Social media post | 1024x1024 | vivid | standard |
| Blog header | 1792x1024 | natural | hd |
| Phone wallpaper | 1024x1792 | vivid | hd |
| Logo / icon | 1024x1024 | natural | hd |
| Quick concept | 1024x1024 | natural | standard |
5. Present for confirmation
Show the user:
- The crafted prompt (exact text that will be sent)
- The chosen size, style, and quality
- A brief explanation of what to expect
Wait for approval before generating.
6. Learn from feedback
After the user sees the result:
- If they like it, save the prompt pattern as a procedural memory
- If they want changes, note what to adjust and iterate
- If they express a style preference, save it to semantic memory
MUST
- Always use the think tool to plan the prompt before presenting it
- Always present the prompt for confirmation before generating
- Always recall style preferences from memory before crafting
- Include an art style specification in every prompt
- Choose parameters that match the stated use case
MUST NOT
- Generate without showing the prompt first
- Use vague prompts ("a nice picture of a sunset")
- Name specific living artists in prompts
- Include text-heavy prompts (DALL-E renders text poorly -- warn the
user if they want text in the image)
- Ignore previously saved style preferences without reason
