Home / Image to prompt: describe an image

Image to prompt: describe an image

The AI looks at your image and writes a description you can use directly as a prompt for Stable Diffusion, Flux or Midjourney.

The AI runs on your device. The first time, your browser downloads the AI model; after that it is faster. On older phones the processing can take a while.
Drop an image here

or tap to choose. Photos, graphics, AI images.

Choose file

Step by step

  1. Drop an image. Photos, graphics and AI images all work, including HEIC photos from an iPhone.
  2. The description starts right away at the selected detail level. The very first time, your browser loads the AI model first; a bar shows the progress.
  3. If the text is too long or too short, set “Detail level” to “Short”, “Medium” or “Detailed” and tap “Describe image” again.
  4. Take the text with “Copy” and paste it into your image generator.
  5. Add what a plain image description doesn't contain: style, lighting, camera angle, aspect ratio or quality terms, depending on what your generator understands.

What you can use it for

  • Starting point for variations: You want to see a subject in other colors, at another time of day or in a different style. The description gives you the base; you only change the parts that should differ.
  • Check what an AI recognizes: Before using an image as a reference, see which things the AI actually picks up. If the main subject doesn't show up in the description, image generators will probably struggle with it too.
  • Sort ideas and references: For a collection of reference images you get a short text for each one that you can search or copy into your notes.
  • Draft an image description: If you have to write about images, for a presentation or a catalog for instance, you get a factual first draft that only needs adjusting.

What is it for?

  • You saw an image you want to recreate, but the prompt is missing.
  • Writing captions for training your own LoRAs.
  • Alt text for websites and online shops.

If the original prompt is still in the file, Read an AI image is more accurate.

Frequently asked questions

Is my image uploaded?

No. The Florence-2 AI model runs in your browser.

Why does it take longer the first time?

Your browser first downloads the AI model. After that it is cached.

Which generators can use the prompt?

Stable Diffusion, Flux, Midjourney and most other image generators.

Which detail level should I use for what?

“Short” gives roughly one sentence, good for captions and quick overviews. “Medium” describes the main elements. “Detailed” is the default and works best as a prompt.

Does the AI recognize people, artists or brands?

The description sticks to what is visible: subject, colors, surroundings, clothing. Don't expect names of people, artists or specific models.

Do I get a negative prompt as well?

No. The tool describes what can be seen in the image. What should not appear, you enter yourself in your generator.

Can I describe many images at once?

This tool describes one image at a time. For many images with captions saved as text files, use the “LoRA dataset preparation” tool.

Is my image analyzed at full size?

For the AI, the image is reduced to at most 1024 pixels on the long side. That is enough for a description, but the finest details such as small text are more likely to be missed.

Why is the result sometimes inaccurate?

The model is small enough to run in a browser. With unusual subjects, abstract graphics or lots of detail it may miss things or name them wrongly, so read the text through once.