Chrome extension · v1.0.0

Every image is a prompt you have not read yet.

Right-click any picture on the web and Unprompter writes the prompt that would recreate it — subject, light, lens, palette, mood, and what to avoid. One reading, nine prompt formats, and the image goes straight to your own model. No credits. No middleman.

Get Pro — $15 once See how it works Free tier: 10 images a day, no card
Nine prompt formats from one reading Your API key, or Ollama on your own machine EXIF stripped before anything is sent
any-website.com/gallery
Unprompt this image
measured 3:2 · warm, low key
midjourneycopied ✓
// one reading → nine formats
a lone standing stone on a moorland ridge,
heather and wet granite underfoot, distant
rain squall, low winter sun raking from
camera left, long shadows, 35mm, f/8,
deep focus, warm, natural, low key,
amber, deep violet, charcoal, documentary
landscape photography, 35mm film grain
--ar 3:2 --style raw --no people, roads --v 7
Writes for Midjourney Stable Diffusion Flux DALL·E Ideogram Gemini Sora / Veo Plain English JSON

Three steps

Right-click. Read. Paste.

No account, no upload queue, no credit balance. The picture goes from your browser to the model you already pay for, and the prompt comes back.

1

Point at anything

Right-click an image, hover it and hit the badge, drag a box round part of the screen, or drop a file into the panel. Screen regions work on canvases, video frames and CSS backgrounds — anything you can see.

2

One reading

Your model returns a structured reading: subject, scene, composition, camera, lighting, colour, texture, mood, style, any lettering, and a negative list. The palette is measured off the pixels rather than guessed.

3

Nine ways out

Switch format and the same reading is rewritten for Midjourney flags, SDXL weights, a Flux sentence, a Gemini instruction, a video shot with a camera move. Switching costs nothing — no second API call.

What is inside

Built for people who actually ship images.

Nine prompt dialects

Every generator wants a different shape. Midjourney takes stacked phrases and flags; SDXL wants weights and a separate negative; Flux wants a sentence. One reading feeds them all.

The nine formats →

Your model, your bill

OpenAI, Claude, Gemini, OpenRouter, or Ollama and LM Studio running on your own machine. Local means the image never leaves the building, and every reading is free.

Pick a provider →

Nothing passes through us

There is no Unprompter server in the path. EXIF and GPS are stripped before upload, history stays on your machine, and site access is asked for one origin at a time.

What leaves your machine →

Variations without a second call

Same shot at night. Same subject, pulled wide. Same scene as an illustration. Eight one-tap remixes rewritten locally from the reading you already paid for.

Borrow one look for another subject

Take the subject from your product shot and the lighting, grade and finish from an ad you admire. Unprompter blends the two readings into a single prompt.

A whole page at once

Point it at a moodboard, a competitor's gallery or a Pinterest board and read every image on the page into a searchable, exportable library.

The honest bit

A prompt is a reading, not a recovery.

Unprompter does not retrieve the original prompt — nobody can, it is not in the file. It reads the picture and writes instructions that land in the same place. On style, palette and composition that gets you very close. On a specific face or a brand's exact typeface it will not, and the tool says so rather than inventing a name.

Fifteen dollars. Once.

Three browsers, lifetime updates, no subscription and no credits to top up. The free tier stays free at 10 images a day.