ComfyUI-InfiniDepth

ComfyUI-InfiniDepth

ComfyUI custom nodes for InfiniDepth — arbitrary-resolution monocular depth estimation (CVPR 2026).

Nodes

| Node | Description | |------|-------------| | Load InfiniDepth Model | Load depth estimation model from checkpoint | | InfiniDepth Depth Estimation | Run depth estimation on images, outputs raw + colorized depth maps |

Installation

1. Clone this repo into ComfyUI's custom nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/snomiao/ComfyUI-InfiniDepth.git

2. Clone the InfiniDepth repo and set the environment variable:

git clone https://github.com/YvanYin/InfiniDepth.git
export INFINIDEPTH_ROOT=/path/to/InfiniDepth

Or symlink this node folder into the InfiniDepth repo (auto-detects InfiniDepth/ as sibling directory).

3. Install dependencies:

pip install -r requirements.txt
python install.py  # installs MoGe and gsplat from git

4. Download checkpoints from HuggingFace and place them as:

checkpoints/
  depth/infinidepth.ckpt
  depth/infinidepth_depthsensor.ckpt
  moge-2-vitl-normal/model.pt
  gs/infinidepth_gs.ckpt
  gs/infinidepth_depthsensor_gs.ckpt
  sky/skyseg.onnx

Usage

1. Add Load InfiniDepth Model node — select model type and checkpoint paths
2. Add InfiniDepth Depth Estimation node — connect model + image
3. Connect outputs to Preview Image nodes

Outputs:

• depth_raw: Grayscale depth map (normalized 0-1)
• depth_colored: Spectral colormap visualization

Supported Platforms

• CUDA (recommended)
• MPS (Apple Silicon)
• CPU (slow but works)

Credits

• InfiniDepth by Tianyu Li et al.
• MoGe-2 by Microsoft for metric scale recovery