Alt text generator without upload
Have images described automatically – as short alt text for websites and social media, or as a detailed description. The vision model (Florence-2) is downloaded once and then runs on your device: your images are not uploaded anywhere. Several images at once, output as text, alt attribute or CSV.
Florence-2 (Microsoft, MIT) · one-time ≈ 190–215 MB ✓ stored
Pick an image on the left – the description appears here and can be edited straight away.
How it works
Choose images
Drop one or more images on the field, pick them, or paste an image from the clipboard.
Describe locally
The first time, the browser downloads the vision model (about 190–215 MB) and keeps it. After that an image takes about a second with a graphics card, otherwise 15–30 seconds.
Check & use
Pick the length, touch up the text if needed and copy it – as plain text, as a ready-made alt attribute, or for all images as a CSV table.
Frequently asked questions
Are my images uploaded?+
No. The image is analysed in your browser and does not leave your device. The only downloads are the vision model (from Hugging Face) and the runtime (from jsDelivr), once; for German text possibly a translation model as well.
What is alt text?+
Alternative text describes an image for people who cannot see it – screen readers read it aloud, and it is shown when the image fails to load. Search engines use it too. Good alt text is short (up to about 125 characters), names what matters and does not start with “image of …”.
How good are the descriptions?+
The model usually gets the subject, people, colours and setting right. It knows neither names nor the context of your page, does not always count correctly and can invent details. Treat the result as a draft: read it once and add what matters to you about the image.
Why is the German text sometimes clumsy?+
The vision model describes in English. German text is produced by an automatic translation on your device – with the browser’s built-in translator or with Opus-MT (about 108 MB, once). That occasionally gives unusual wording; you can change the text right in the field.
Which images work?+
Photos, product shots, illustrations and screenshots in JPG, PNG, WebP, GIF and AVIF. For text inside images the OCR tool is the better choice; charts and tables are only described roughly.
Does it work offline?+
Yes, once the models have been downloaded. They stay in your browser’s storage; “Delete stored models” frees the space again.
Can I save the description into the image file?+
Can the tool give keywords too?+
Yes – with “Add keywords” a list such as “street, people, tent, dog” appears below the description. The model looks for the nouns of its short description in the picture and adds things it detects that the text leaves out. You can edit and copy the list, get it as a column in the CSV table, and write it into the image file as keywords (XMP dc:subject) when saving. It costs two more runs per image – about one to three seconds with a graphics card.
Image description with Florence-2 base (Microsoft, MIT licence) via transformers.js; German output via the browser translator or Opus-MT (Helsinki-NLP, CC BY 4.0) – entirely on your device. Automatic descriptions may contain mistakes and should be checked before publishing.