Is it free?
Image analysis is free, with no account or credits. Repeat analyses are welcome, subject to fair-use request limits and temporary service capacity. See free access.
Free · No account · No credits
Image to Flux prompt generator
Upload a reference image and get a Flux prompt written as complete natural-language sentences, with the spatial relationships between foreground, subject, and background stated instead of implied.
01
Add a reference
Describe the visible image closely.
Free · No account · No credits
Resized and stripped of metadata before analysis
02
Your prompt
A real reference-to-prompt case
The scene carries over. Exact shape and framing can change.
AI-generated demo images · Opening this example makes no analysis call.See the full case and prompt formats ↗Why Flux gets sentences instead of tags
Flux responds to prose. The compile step therefore writes clear natural-language sentences with explicit spatial relationships, and avoids tag spam and excessive negative phrasing — the opposite of the comma-separated concept lists used for diffusion-style targets.
That only works if the underlying facts are right. A vision pass first separates the scene into foreground, middle ground, and background, records framing, camera, lighting direction, palette, and materials, and quotes the exact readable text. Facts it cannot resolve are reported as uncertainties instead of being filled in.
A second vision pass then compares the draft with your image and scores it on six components, including a model-fit score for the target you chose.
What a Flux result contains
- A primary prompt in full sentences.
- An alternate wording that stays faithful to the same image.
- A negative prompt, filled only where the target benefits from one.
- A structured breakdown covering subject, environment, composition, camera, lighting, palette, materials, style, mood, visible text, and parameters.
- A quality report with weighted component scores.
The prompt is editable in place, so you can tighten a sentence or drop a clause before copying it into Flux.
Faithful or Balanced
Faithful mode prioritises literal recreation of the visible image. Balanced mode keeps the subject and composition while allowing restrained polish in the wording. The review pass runs either way.
Worked example: travel street photo
A red scooter parked on a narrow cobblestone street in Lisbon at golden hour. Tram tracks curve into the background, laundry hangs from a balcony above, long warm shadows cross the frame.
Primary prompt
A red vintage scooter is parked on a narrow cobblestone street in Lisbon during golden hour. Tram tracks curve from the foreground into the soft-focus background, and laundry hangs from a tiled balcony above the scene. Long warm shadows stretch across the street, with the low sun behind the buildings creating a gentle rim light on the scooter's chrome details.
Flux responds to complete natural-language sentences with explicit spatial relationships — foreground tracks, mid-frame scooter, background balcony — instead of comma-separated tag spam.
Three more use cases are written out on the image to prompt examples page.
Frequently asked questions
Why is there no negative prompt for Flux?
The compile step avoids excessive negative phrasing for Flux, so the field is left empty rather than filled with exclusions the target does not need.
How long can the prompt be?
There is no fixed length. The prompt is as long as the verified observations require, and the structured breakdown lets you see which fact each part came from before you trim it.
Are my images stored?
No. Your browser resizes the image and strips file metadata before upload, the prepared image is analysed in memory for the duration of the request, and this site does not save it.
Other target models
The same pipeline compiles for a portable universal format as well as the other image models. Each page preselects its own target.
- Universal image to prompt
- Midjourney image to prompt
- Stable Diffusion image to prompt
- DALL·E image to prompt
Read the full four-step pipeline on how image to prompt works.