CoolFace
Apppublic

galexdav/fastvlm-vision

sourceHugging Faceupdated 4mo agoView on Hugging Face
0likes
App README

👁️ FastVLM Vision

Real-time vision analysis powered by [FastVLM-0.5B](https://huggingface.co/apple/FastVLM-0.5B) — running 100% locally in your browser. No server, no API key, no data ever leaves your device.

How it works

Point your laptop camera at anything and ask FastVLM what it sees. The model runs entirely on your GPU via WebGPU (with automatic CPU fallback), so inference is fast and private.

  • —Analyze Frame — captures the current camera frame and streams a description
  • —Auto mode — re-analyzes every 6 seconds, great for watching a scene change
  • —Custom prompt — change the question to anything: "What objects are on the desk?", "Is there text visible?", "Describe the mood of this scene."

First load

The model weights (~700 MB, quantized to 4-bit) download to your browser's cache on first visit. This takes a minute or two depending on your connection. Every subsequent visit loads instantly from cache.

Browser support

BrowserStatus
Chrome 113+ / Edge 113+✅ Full WebGPU
Firefox 141+ (Windows)✅ WebGPU
Safari 26 (macOS / iOS)✅ WebGPU
Older browsers⚠️ Falls back to CPU (slower)

Model