Visual Describer
The flagship. Splits your image into quadrants and describes each region for more thorough results. Works in light and dark mode.
Main describers
Drop or paste an image, get alt text. Start here.
Alt generator — tuned
The original single-page generator — click, drop, or paste an image and get alt text back. Fastest option if you just need a description.
Alt Text Next
A preview of the next major version. Functionally complete, still being polished.
Offline generator
Runs entirely in your browser, saves nothing to a server. Bring your own API key and nothing leaves your machine except the image going to the model.
Advanced tools
Specialised tools for when plain alt text isn't enough — deeper analysis, longer critiques, safety scans, and model comparisons.
The Art Critic
Critiques an image as art — composition, mood, technique — then gives a subjective Hot Take. Great for artwork and photography; overkill for screenshots.
Safe Spaces
Scans an image for unsafe content and flags what it finds by category. Meant as a moderation aid, not a gatekeeper.
Multi-scan describer
Runs the image through three passes — overview, detailed scan, and final synthesis — then merges them into one description.
Parallel scan
The three-pass describer but all passes run at once. Faster, uses more API quota.
Sequential describer
A three-stage pipeline where you can pick a different model for each stage — overview, detailed analysis, and synthesis. Good for mixing a cheap model with a precise one.
Variant describer
Runs the same image through several models side by side so you can compare their outputs and pick the one you like.
LM Studio local backend
Uses your own LM Studio server as the vision backend. Keep everything on your machine, use whichever model you've loaded locally.
Pick a specific model
Same simple describer, one vision model each. Use whichever you have an API key for, or try a few and compare how they describe the same image.
xAI Grok Vision
Grok's vision model. Fast, conversational, decent on photographs.
Claude (Anthropic)
Claude's vision capability. Tends toward careful, thorough descriptions.
GPT-4 Vision (OpenAI)
OpenAI's vision. Good general-purpose baseline — often the fallback if another model struggles.
Pixtral (Mistral)
Mistral's Pixtral vision model. Experimental — sometimes excellent, occasionally patchy.
Pixtral — saved history
A variant of the Pixtral describer that keeps a history of images and their generated alt text.
Pixtral — scratch
A smaller Pixtral sandbox for quick experiments. Same model, stripped-down UI.
Hugging Face vision
Uses Hugging Face Inference for a smaller, open vision model. Cheaper and slower than the commercial ones above.
Camina (via Coze)
A describer routed through a Coze bot tuned for accessible, ethical image descriptions. Slower, more careful output.
Gemma 3 (not working)
Google's Gemma 3 vision variant. Currently broken — kept here so the work isn't lost.
Work in progress
Development copies kept live so they can be tested in the open. Expect rough edges.
Dev mirror — Visual Describer
Bleeding-edge copy of the main Visual Describer. Same tool, newer code.
Dev mirror — Alt generator
Dev copy of the tuned single-page generator.
Modular rebuild test
A rebuild of the alt generator with the UI split into smaller reusable pieces. For testing, not production use.
Original intermission page
The "be right back" page that used to live at /alt/. Kept for continuity.
Desktop apps & source code
If you'd rather not use a web tool, these run on your own computer. Linked to their downloads or GitHub repositories.
Alt Text Generator — desktop
Downloadable desktop app. Runs a local vision model so your images never leave your computer.
alt-text-local-llm — source
Source code for the desktop app. Electron + Ollama. Clone and build your own.
Accessible Image Viewer — source
A full image viewer with EXIF parsing, Imgur integration, and alt-text generation built in. Private repo — ask for access.
Bluesky Alt Text Dataset — source
The scraper that built the dataset above. Collects and validates alt text from the Bluesky firehose.