Three options, in order of speed → reliability:

1. ComfyUI Manager (recommended) — search for Depth Anything V3 in the Manager and click Install from the highest version displayed. If that doesn't work, try nightly.
2. Manager via Git URL — in ComfyUI Manager: "Install via Git URL" with https://github.com/PozzettiAndrea/ComfyUI-DepthAnythingV3.git.
3. Manual (most reliable):

   cd ComfyUI/custom_nodes
   git clone https://github.com/PozzettiAndrea/ComfyUI-DepthAnythingV3.git
   cd ComfyUI-DepthAnythingV3
   pip install -r requirements.txt --upgrade
   python install.py

Please report any problems you hit during installation or use of my nodes — open a Discussion or Issue. Very grateful for your help! 🙏

View Live Test Gallery →

Custom nodes for Depth Anything V3 integration with ComfyUI.

Simple workflow:

Advanced workflow:

Single image to 3d:

Multiple image to 3d:

Single image to mesh:

Use multi attention node for smooth video depth!

You can use the multi-view node to use the cross attention feature of the main class of models. This is done to have a more consistent depth across frames of a video.

You can reconstruct 3D point clouds!

Even from multiple views, with the option to either match them (with icp) or leave them to use the predicted camera positions. You also have a field on the point cloud to show you which view each point came from.

Depth Anything V3 is the latest depth estimation model that predicts spatially consistent geometry from visual inputs.

Published: November 14, 2025 Paper: Depth Anything 3: Recovering the Visual Space from Any Views

Different models support different features:

• ✅ = Fully supported
• ❌ = Not available (returns zeros/ignored)
• ⚠️ = Works but no cross-view attention benefit (images processed independently)
• ✅* = Requires fine-tuned model weights (placeholder in current release)

Choose your model based on needs:

• Need sky masks? → Use Mono/Metric/Nested (required for V2-Style normalization)
• Need camera conditioning? → Use Main series or Nested
• Processing video/multi-view? → Use Main series or Nested for consistency
• Single images only? → Any model works

1. Use Mono or Metric models (they provide sky segmentation)
2. Set normalization_mode to V2-Style (default)
3. Connect the depth output to your ControlNet node
4. Enjoy clean depth maps with proper sky handling!

1. Use any model (Mono/Metric recommended for sky filtering)
2. Set normalization_mode to Raw
3. Connect depth → depth_raw, confidence → confidence, sky_mask → sky_mask to DA3 to Point Cloud
4. Sky pixels will be automatically excluded if sky_mask is connected
5. Important: Point cloud nodes validate input and will raise an error if normalized depth is detected (prevents incorrect 3D output)

Questions or feature requests? Open a Discussion on GitHub.

Join the Comfy3D Discord for help, updates, and chat about 3D workflows in ComfyUI.

• Original Paper: Haotong Lin, Sili Chen, Jun Hao Liew, et al. (ByteDance Seed Team)
• Original Implementation: PozzettiAndrea
• V2-Style Normalization: Ltamann (TBG) - See TBG Takeaways: Depth Anything V3 for workflow examples
• Based on: Official Depth Anything 3 repository
• Inspiration: kijai's ComfyUI-Depth-Anything-V2

Model architecture files based on Depth Anything 3 (Apache 2.0 / CC BY-NC 4.0 depending on model).

Note: Some models (Giant, Nested) use CC BY-NC 4.0 license (non-commercial use only).