About

You found an image you like. Writing the prompt that recreates it is the hard part. img2prompt does that part, in the syntax your model expects.

It is for people who work from references: illustrators building a moodboard into a set, product teams matching a shoot they already have, filmmakers storyboarding with Sora or Kling, and anyone who has spent an evening guessing which words made someone else's image look the way it does.

We built it because every model reads prompts differently. A Midjourney prompt full of parameters means little to DALL·E, and a Stable Diffusion prompt without weights and a negative prompt leaves results on the table. One image should give you a prompt that fits each of them.

What it does with one photo

One upload, read once, written out for each model. Example output.

Open the tool
Overhead photo of a cup of coffee on a wooden café table in morning light
cafe_morning.jpg
Prompts
Midjourney
overhead shot of a cup of black coffee on a worn wooden café table, soft morning window light, warm brown tones, quiet slow-living mood, editorial food photography --ar 4:3 --style raw --v 7
Stable Diffusion
(cup of coffee:1.2), overhead view, wooden table, (morning window light:1.1), warm tones, café, editorial photo, highly detailed Negative prompt: blurry, lowres, text, watermark
DALL·E 3
An overhead photo of a cup of black coffee on a worn wooden café table, lit by soft morning light from a window. Warm brown tones and a calm, unhurried mood.
Kling
Static overhead shot, 5s. Steam rises slowly from a cup of coffee on a wooden table as morning light shifts across the grain.

Subject, light, palette and style are read once. Only the wording changes per model.

What we believe

A prompt is a starting point

No prompt recreates an image pixel for pixel, and we do not pretend it does. The tool gives you an accurate, editable draft that gets the subject, light and style right, so your time goes into the parts that are yours.

Write for the model, not for the reader

A prompt that reads nicely is not the same as one a model follows. We keep each model's syntax current, from Midjourney parameters to Stable Diffusion weights to the camera language Sora and Kling respond to.

Your images are not our data

An upload exists to produce a prompt. It is never shown to anyone else and never used to train a model, and the full-size image is not kept once the prompt is written.

14

Models with their own prompt syntax

8

Languages prompts can be written in

50

Images in a single batch

30

Days a history thumbnail is kept

How we treat your images

We think an image tool should be boring about privacy. When you upload an image, it is sent over an encrypted connection to the AI model that writes the prompt. We do not keep the full-size image, we do not build a dataset from uploads, and we do not use them to train or tune any model.

Your history keeps the prompts and a small thumbnail so you can recognise each entry. Thumbnails are deleted after 30 days. You can delete any entry, or your whole account, from the dashboard. The full details are in our privacy policy.

Upload
Sent over HTTPS to the model that writes the prompt
After analysis
Full-size image discarded. Never used for training
Kept
The prompts, plus a small thumbnail for 30 days

Questions, ideas, partnerships

We read every message and reply as soon as we can.