Depth Anything V2 runs in your browser

Monocular depth estimation via Depth Anything V2 (Yang et al., 2024) · powered by Transformers.js + ONNX Runtime · no server, no GPU needed

Model size Small (25M — fastest) Base (98M — balanced) Large (335M — best quality)

Device WebGPU (if available) WASM (universal)

  Compute Depth

🖼️

Click to upload or drag & drop an image

JPG, PNG, WebP — max ~4K recommended

Or try an example:

Initializing…

⬇ Download depth map (PNG)

First run downloads the model (~25–335 MB depending on size) and caches it. Subsequent runs are instant to load.