Describe Image with AI

It is a vision-language model connected to a simple interface: you upload a picture, it returns English text about what is visible. The useful part is not the raw capability — plenty of models can do this — but having it packaged so the output lands in the right shape. Instead of one generic paragraph every time, you pick the shape you need: a description, a one-line summary, an object list, a text transcription, a style analysis, alt text with a caption, or a direct answer to a question you typed in.

For well-lit pictures with a clear subject, the objects, setting, colours, materials and spatial relationships are reliably described, and prose of four to eight sentences usually needs no correction beyond emphasis. Where accuracy drops off is predictable: exact counts in crowded frames, small text, anything behind glass or in a reflection, heavy filters, and identifying a specific real person. Anything that will be published, billed or relied upon should get a human read before it goes out.

JPG, PNG and WEBP up to 20 MB, with both sides at least 256 pixels. HEIC files straight from an iPhone are accepted too and converted to JPEG inside your browser before upload — the picture itself never leaves your device during that conversion. Animated PNG and animated WEBP are rejected because there is more than one frame to choose from.

Use Length below the modes. Short keeps every answer to one compact sentence, Balanced gives two to four sentences, and Detailed writes five to eight sentences covering setting, materials, colour, light and spatial relationships. Length applies to whichever mode you pick.

Yes. Select Ask a Question, type up to 200 characters in the box underneath — how many windows are in this room, is there something in the dog's mouth, what colour is the car — and you get a plain answer drawn from the picture. If the frame does not contain enough information, the answer says what is missing rather than inventing something.

Yes. Extract Text transcribes whatever is legible in the frame — screenshots, receipts, signage, slides, menus, name badges, documents — keeping the original language and line breaks intact, so it works for Chinese, Spanish or German sources and can be pasted straight into a translator. Text needs to be reasonably sharp; anything below roughly twenty pixels in height tends to come back as gibberish or get skipped.

Both. Alt Text and Captions is built exactly for accessibility: alt text under 125 characters written to be read aloud, a caption of twenty words or fewer, and three to six lowercase keywords. For listings, run Recognize Objects together with Ask a Question — one gives you the prose for the description field, the other the itemised list for the specification block. Check any claim about material, origin, safety or compliance yourself before publishing.

Your source image and description are saved to your account history so you can reopen them together. Sign in to access your history.

Different output for a different job. This page writes for people — complete sentences you can paste into alt text, a product description or a caption. An image to prompt tool writes for models: compact tags, weighted terms and parameters such as aspect ratio and stylize value, intended to be fed into Midjourney, Stable Diffusion or Flux to generate something new. If you have finished describing the picture and now want to turn those words into a reusable prompt, open the image to prompt tool and start there.

Each description costs 3 credits and requires an account. Failed or timed-out requests are refunded.