Skip to content
FastVLM.net

FASTVLM-0.5B · TRANSFORMERS.JS · WEBGPU

FastVLM Playground

Ask questions about a photo or screenshot, extract text, or explain a chart. Run FastVLM on your device and export the answer with measured timings.

In-browser inference Images + camera No local install

FASTVLM · IMAGE PLAYGROUND

Ask about your own image

Upload a photo, paste a screenshot, or start with a sample. FastVLM-0.5B runs on your device.

Checking this browser before downloading…

JPEG, PNG or WebP · up to 10 MB · Ctrl/Cmd+V to paste

The first run downloads about 1.1 GB of model files from Hugging Face. Later runs can reuse your browser cache. Start when you are ready.

Run locally with Python →

Answer

Your answer and measured timings will appear here.

Images are resized to a maximum of 1024 px on the longest side. Generated answers can contain mistakes.

Images, questions and answers stay on your device. Usage events record steps, sample/upload/paste input type, errors and timings.

onnx-community/FastVLM-0.5B-ONNX
Inspect 24 image tests, including Chinese failure cases and full answers →
Image questions

Browser requirements

  • A current desktop version of Chrome or Edge with WebGPU enabled is the safest option.
  • A WebGPU-capable GPU and enough free memory for the model and browser tab.
  • Camera permission is only needed for live-camera input; allow it when the embedded demo asks.
  • The initial model download is much slower than later runs and depends on your connection.

What model does this use?

The browser playground targets FastVLM-0.5B, the smallest FastVLM variant. Apple links this real-time Transformers.js + WebGPU experience from its official research page.

Apple Research

Privacy and data flow

Model inference is designed to happen in your browser. FastVLM.net does not receive your camera frames. The embedded Hugging Face application and your browser still handle standard requests, model downloads and permissions; review their policies for complete details.

Troubleshooting

The demo stays blank

Wait for the iframe and model assets to load, then try opening the demo directly. Disable strict content blockers for the embedded application if needed.

WebGPU is unavailable

Update Chrome or Edge, confirm hardware acceleration is enabled, and check that your GPU is not blocklisted by the browser.

Camera does not start

Allow camera access for both fastvlm.net and the embedded Hugging Face app, then reload the page.

The first run is slow

This is expected while model weights download and WebGPU shaders compile. Later visits can be faster when browser caching is available.

See the playground in action

A short recording of the FastVLM camera experience. Open the camera demo above to try it yourself.

Continue building with FastVLM

Compare model sizes or move from the browser playground to a reproducible Transformers workflow.

Compare models Read Transformers guide