Depth Anything 3: Monocular Depth Estimation
Depth Anything 3 by ByteDance Seed provides BASE, Small, Metric Large, and Mono Large variants for monocular depth estimation in ComfyUI.
Depth Anything 3 is a monocular depth estimation model from ByteDance Seed. It enables high-quality depth map generation from a single input image, useful for conditioning ControlNet, generating 3D effects, and image compositing in ComfyUI workflows.
Variants
| Variant | Description |
|---|---|
| BASE | General-purpose depth estimation |
| Small | Lightweight, faster inference |
| Metric Large | Metric (absolute) depth estimation |
| Mono Large | Relative/ordinal depth with higher quality |
Usage in ComfyUI
Place the model files in ComfyUI/models/geometry_estimation/. Requires a ComfyUI custom node that supports Depth Anything 3 for loading and inference.
| Developer | ByteDance Seed |
| Release Date | 2025 |
| Architecture | Monocular Depth Estimation Transformer |
| License | Apache-2.0 |
| Type | Geometry Estimation / Depth |
| Upstream | https://github.com/ByteDance-Seed/Depth-Anything-3 |
Links
- Hugging Face (Comfy-Org): Comfy-Org/Depth-Anything-3
- Upstream: depth-anything/DA3-BASE
- GitHub: ByteDance-Seed/Depth-Anything-3
Guides and workflows related to this model series.
No articles found.