Ask FastVLM about an image

Choose a photo, paste a screenshot or try a sample. The 0.5B model runs on your device and streams its answer here.

Checking browser support…

Your image

Processed on your device

Choose image

JPEG, PNG or WebP · up to 10 MB. Paste a screenshot while focused in this panel. Images are resized to a maximum of 1024 px on the longest side.

Choose an image or try a sample

Your question

The first run downloads about 1.1 GB from Hugging Face. Later runs may reuse this browser’s cache. Model preparation also needs free GPU memory.

Answer

Choose an image and ask a question.

Your answer will appear here. Check generated text against the original image before using it.

Timings describe this run on your device. “First text” includes image processing and is not a strict first-token benchmark. JSON includes the question, answer and timings, without image pixels.

Run locally with Python →

This image playground processes your image and question locally. Model files are downloaded from Hugging Face and runtime files from jsDelivr. Usage events record action categories and timings, without images, file names, questions or answers. Browser model: onnx-community/FastVLM-0.5B-ONNX.

Try the camera demo

Apple’s separate Hugging Face demo supports a live camera feed. Opening it connects to Hugging Face; camera access is requested inside that demo.

Open in a new tab

Build this into your own app with the WebGPU starter, or compare the available models and runtimes.

Inspect the original site's 24-image evaluation and failure cases, or benchmark your device.