Prompt Detective
Where Image to Prompt describes a picture in prose, this returns a structured blueprint: quality tags, subject, setting, and camera as separate fields. That separation is the whole point.
Why structure beats a paragraph
A single block of text has to be rewritten wholesale for every change. Four labelled parts let you edit one and leave the rest alone.
- Keep the setting, swap the subject.
- Keep the subject, change the camera.
- Strip the quality tags entirely and see what actually changes. Usually less than you expect.
A workflow that teaches you something
Run an image you admire through the detective. Replace only the subject line with your own character. Render.
What survives
Whatever still looks like the original was coming from the setting and camera, not the subject.
What vanishes
Whatever disappears was carried by the specific character description.
Why it matters
You now know which half of any prompt is doing the work - A lesson that transfers to every prompt you write afterwards.
What it cannot recover
The seed and the exact model are unknowable from an image alone. You will get close in register and composition, never an exact match. The blueprint guide explains why order matters as much as wording.
Worked example: keeping a scene and swapping the character
| Input | An image whose mood you want |
|---|---|
| Returns | Four editable fields |
| Cost | 5 credits |
| Test | Change only the subject line |
The exact instruction
quality: masterpiece, best quality | subject: [replace this with your character] | setting: rain-lit balcony at night, neon from below | camera: medium shot, eye level
What comes back
Render it twice: once as returned, once with your own subject. Whatever stays the same was coming from the setting and camera. That tells you which half of any prompt is doing the work.
When it goes wrong
| What you see | Why | What to change |
|---|---|---|
| The rebuild looks different | Seed and model are unknowable from an image | Expect the register to match, not the picture |
| Fields came back merged | Source had an ambiguous composition | Crop to a single clear subject and retry |
| I want plain prose | Wrong tool for that | Use Image to Prompt |
Common questions
How is this different from Image to Prompt?
That returns a description. This returns a structured, editable prompt split into parts you can modify independently.
Will the blueprint recreate the original?
It gets close in register and composition. Not exactly, because the seed and model are unknown.
Can I use it on my own images?
Yes. It is a good way to recover a prompt you forgot to save.
Where to go next
This tool is usually used alongside prose description instead, run the blueprint and prompt structure explained.