AltText Gen
Right-click any image → generate descriptive alt text. On-device ML, no servers, no API keys.
Right-click any image → 'Generate Alt Text' — instant descriptive caption
On-device ML (MobileNet v2, 13MB) via LiteRT.js — 100% offline, no API keys
Top-5 image classification with confidence scores
Template-based caption generation from classified labels
Copy to clipboard or set as the image's alt attribute directly
Editable result — refine the generated text before applying
Stats: total images described
100% free, no premium, no tracking, no servers
AltText Gen generates descriptive alt text for any image — right-click, select “Generate Alt Text”, and get an instant caption. Powered by MobileNet v2 on-device, no servers, no API keys, no data leaving your browser.
How it works
- Right-click an image → context menu shows “Generate Alt Text”
- Image is processed — resized to 224×224, normalized to [-1, 1] range
- MobileNet v2 inference — LiteRT.js runs the model (WebGPU preferred, WASM fallback)
- Top-5 predictions — e.g., “golden retriever (0.89), tennis ball (0.31), dog (0.28)”
- Caption generated — template-based: “A golden retriever” or “Image of a golden retriever with tennis ball”
- Result popup — editable text, “Copy” and “Set as alt” buttons
The ML model
| Property | Value |
|---|---|
| Model | MobileNet v2 |
| Size | 13.3 MB |
| Input | [1, 224, 224, 3] float32 |
| Output | [1, 1000] float32 (ImageNet classes) |
| Runtime | LiteRT.js (WebGPU/WASM) |
| Inference time | 100-500ms |
MobileNet v2 is a pre-trained image classification model that recognizes 1000 ImageNet categories — animals, objects, food, landmarks, vehicles, scenes, and more.
Caption generation
The model outputs class probabilities, not natural language. AltText Gen uses template-based caption generation:
| Predictions | Generated caption |
|---|---|
| golden retriever (0.89) | “A golden retriever” |
| golden retriever (0.85), tennis ball (0.40) | “A golden retriever with tennis ball” |
| pizza (0.92) | “A pizza” |
| laptop (0.78), desk (0.35) | “A laptop on desk” |
The result is always editable — you can refine the text before copying or applying it as the image’s alt attribute.
Privacy
AltText Gen runs entirely on-device. The ML model is bundled with the extension
(13MB). No images, no data, no telemetry is ever sent to any server. The only
permissions are storage (settings + stats) and contextMenus (right-click menu).
Free forever
AltText Gen is 100% free. All features are available to everyone. There are no premium tiers, no API keys, no usage limits. If you enjoy it, you can buy me a coffee — but there’s zero pressure.
How would you rate AltText Gen?
Click a star to vote — no account needed.
Frequently asked questions
Is AltText Gen really free?
Yes, 100% free. All features are available to everyone. No premium tiers, no API keys, no usage limits. If you enjoy it, you can buy me a coffee — but there's zero pressure.
Does AltText Gen send my images to a server?
No. AltText Gen runs entirely on-device. The ML model (MobileNet v2, 13MB) is bundled with the extension and inference happens in your browser via LiteRT.js (WebGPU/WASM). No images, no data, no telemetry is ever sent to any server.
How does the alt text generation work?
AltText Gen uses MobileNet v2, a pre-trained image classification model. When you right-click an image, it's resized to 224x224 pixels, normalized, and fed through the model. The top-5 class predictions (e.g., 'golden retriever', 'dog', 'mammal') are then combined using template-based caption generation to produce a descriptive sentence.
Is this real image captioning or just classification?
It's classification with template-based caption generation — not true neural image captioning (which requires much larger models). MobileNet identifies what's in the image (objects, animals, scenes, food), and templates turn those labels into descriptive text. The result is good for common objects and scenes but won't describe complex compositions.
What kinds of images work best?
AltText Gen works best on photos with clear subjects — animals, objects, food, landmarks, vehicles. It identifies 1000 ImageNet categories. Abstract art, screenshots, or complex multi-subject scenes may produce less useful results.
Will AltText Gen slow down my browser?
No. The model only runs when you explicitly right-click an image and select 'Generate Alt Text'. It doesn't run on page load or in the background. Inference takes 100-500ms depending on your hardware (WebGPU if available, WASM fallback).