What is this?
A small collection of tools for generating image alt text — from a simple drop-and-go web page to multi-pass analyzers, a safety scanner, and an art critic. All of them run in your browser. Some talk to commercial vision models, some talk to a local model on your own machine, and one runs entirely client-side.
Open the gallery, pick a tool, drop an image in. Most tools accept drag-and-drop, paste (Ctrl+V / Cmd+V), or a click to upload.
Which one should I pick?
| If you want… | Use this |
|---|---|
| A quick, accessible description | Alt generator — tuned or the Visual Describer |
| More thoroughness (image split into regions) | Visual Describer |
| A description focused on art and composition | The Art Critic |
| A safety scan of an image | Safe Spaces |
| Several models compared on the same image | Variant describer |
| A specific vision model | Pick a model section |
| A local LM Studio server as the backend | LM Studio backend |
| Everything in your browser, no server | Offline generator |
| A desktop app that runs a model on your computer | Desktop download |
| Alt text for a Tenor GIF | Tenor GIF alt generator |
What do I need?
Most of the web tools need an API key for whichever vision model they use. You paste the key into the page and it stays in your browser's local storage — nothing gets uploaded anywhere except direct calls to the model provider you chose.
- xAI Grok Vision — x.ai/api
- Claude (Anthropic) — console.anthropic.com
- GPT-4 Vision (OpenAI) — platform.openai.com
- Pixtral (Mistral) — mistral.ai
- Hugging Face — huggingface.co/settings/tokens
- LM Studio — nothing external; point the tool at your local LM Studio server
- Desktop app — nothing external; ships with Ollama support for local vision models
A few tools talk to a small proxy on this server instead of calling the APIs directly. If one of those loads but the Generate button doesn't return anything, the proxy is probably offline — try a different tool, use the offline generator, or grab the desktop version.
What happens to my data?
- Your API keys stay in your browser. They're saved to
localStorageand used only for direct calls to the model provider. Clear your browser data and they're gone. - Your images go to whichever model you pick. If you pick OpenAI, OpenAI sees the image. If you pick a local LM Studio model, nothing leaves your machine.
- This site doesn't store your images. The handful of tools that route through the proxy don't keep anything after the response.
- The offline generator and the LM Studio backend are the most private web options. The desktop app is the most private option overall.
Desktop app
If you'd rather keep everything on your own computer, there's a desktop version at dr.eamer.dev/downloads/alt_text/. It bundles an Electron frontend with Ollama for local vision models, so your images never leave your machine.
Source code: github.com/lukeslp/alt-text-local-llm (public).
Bluesky alt-text dataset
A free dataset of 279,000 curated image-description pairs, scraped from 489 validated Bluesky accounts. Good for training, evaluation, or research: dr.eamer.dev/downloads/bluesky-alt-text/.
Reporting a problem or contributing
Something broken? A description looks wrong? Want a new model or feature? Email luke@lukesteuber.com or reach out on Bluesky @lukesteuber.com. The source for this gallery lives in a private repository; if you'd like access for serious contributions, mention that in your email.
Credits
Built by Luke Steuber. The describers use vision models from xAI, Anthropic, OpenAI, Mistral, Hugging Face, Google (Gemma), and Ollama. The Bluesky dataset was collected from the public firehose. Accessibility follows WCAG 2.2 AA.
Site: lukesteuber.com · Bluesky: @lukesteuber.com