Prompt library
Refine an idea into an image-generation prompt
Image models reward specificity in exactly the categories this prompt forces (subject, composition, light, palette, style), and lead-with-subject matches how prompt tokens are weighted. Planning the defect check and the one-parameter iteration turns generation from slot-machine pulls into a controlled search.
Last reviewed July 17, 2026
The prompt
Refine my rough idea into image-generation prompts.
The idea: {{idea}}
Where the image will be used: {{usage}}
Produce:
1. Three refined prompts, each a single paragraph that specifies: subject with concrete attributes, composition and framing (close-up, rule of thirds, negative space for text if the usage needs it), lighting, color palette or mood, style anchor (medium or genre, not artist names), and the aspect ratio for the stated usage.
2. Make the three genuinely different interpretations of the idea (different composition philosophies or moods), not one prompt with synonyms.
3. Per prompt: what to look for in the result to judge it quickly, and the single most likely defect for this kind of request (mangled text, extra fingers, cluttered background) with the phrase that mitigates it.
4. A negative-prompt line for tools that accept one.
5. The iteration move: for each prompt, which ONE parameter to vary on the second attempt if the first is close but off.
Keep every prompt under 120 words; generators weight early words most, so lead with the subject.Run in idaptOpens a new chat with the prompt prefilled. Nothing sends until you press send.
Fill in the variables
| Variable | What it is | Example |
|---|---|---|
| {{idea}} | Your rough idea | something about teamwork for the blog header |
| {{usage}} | Where it goes | blog hero, wide, text overlaid on the left |