Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.comfy.org/llms.txt

Use this file to discover all available pages before exploring further.

​
ComfyUI MoGe Introduction

MoGe (CVPR 2025, from Microsoft Research) is a powerful model for recovering 3D geometry from monocular open-domain images. It estimates metric point maps, metric depth maps, normal maps, and camera FOV — all in a single forward pass. Key capabilities:
  • Accurate 3D geometry estimation: point maps, depth maps, and normal maps from a single image — one model, one forward pass
  • Metric scale (MoGe-2): point maps and depth maps in real-world scale
  • Flexible resolution support: works with various resolutions and aspect ratios (2:1 to 1:2)
  • Fast inference: ~60ms per image on A100 / RTX 3090 (FP16, ViT-L)
  • Mesh generation from both perspective and panoramic images
MoGe comes in two versions:
FileVersionMetric ScaleNormal MapsParams
moge_1_vitl_fp16MoGe-1--314M
moge_2_vitl_normal_fp16MoGe-2✅✅331M
MoGe-2 adds metric scale and high-quality normal map estimation with sharper visual detail and lower inference latency. MoGe-2 with normal (moge_2_vitl_normal_fp16) is the recommended checkpoint.
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup
ComfyUI now natively supports MoGe nodes. Make sure you have updated to the latest version of ComfyUI before starting.Generated .glb format models will be output to the ComfyUI/output/mesh folder.

​
Model Installation

Download the MoGe checkpoint(s) and save them to the corresponding ComfyUI folder: 
ComfyUI/
├── models/
│   ├── geometry_estimation/
│   │   ├── moge_2_vitl_normal_fp16.safetensors
│   │   └── moge_1_vitl_fp16.safetensors

​
Example Workflows

​
1. Depth Estimation

What it does: Takes a single image and estimates its metric depth map, colored depth preview, and mask — outputs the same metric-scale depth that MoGe infers in one forward pass. Useful as a scene depth reference for compositing, depth-based effects, or as preprocessing before mesh generation. MoGe also estimates the camera’s field of view (FOV) from the image, which can be optionally overridden with a ground-truth value for even more accurate results.

Download Workflow

Download JSON or search “MoGe Depth Estimation” in Template Library

Download Sample Image

Get the example input image for this workflow
depth estimation color previewdepth estimation raw preview

​
1.1 Steps to Run

  1. Ensure the LoadMoGeModel node has loaded a MoGe checkpoint
  2. Load an image in the Load Image node
  3. Click Queue or use Ctrl(cmd) + Enter to run
  4. The workflow outputs colored depth preview, raw depth preview, and a mask

​
2. Perspective to Mesh

What it does: Converts a single perspective photo into a textured GLB mesh with normal and depth previews. MoGe estimates point maps, depth, and normals from the visible scene, then converts them to a mesh. This is monocular geometry estimation — occluded areas and object backsides will be missing or fragmented. Useful for quick scene prototyping, reference geometry, or visualizing depth and normals as a mesh, but not a replacement for multi-view 3D reconstruction.

Download Workflow

Download JSON or search “3D MoGe Perspective to Mesh” in Template Library

Download Sample Image

Get the example input image for this workflow
perspective to mesh preview

​
2.1 Steps to Run

  1. Ensure the LoadMoGeModel node has loaded a MoGe checkpoint
  2. Load a perspective photo in the Load Image node
  3. (Optional) View the OpenGL and DirectX normal previews
  4. Click Queue or use Ctrl(cmd) + Enter to run

​
3. Panorama to Mesh

What it does: Converts an equirectangular (360°) panorama into a textured GLB mesh. The workflow uses MoGePanoramaInference to split the panorama into 12 perspective views, runs monocular geometry estimation on each view independently, then merges them into a single mesh. Each segment is still single-view estimation, so the result is a rough scene reconstruction — useful for getting a spatial overview of a 360° scene, but occluded areas and geometry behind surfaces will be missing or fragmented.

Download Workflow

Download JSON or search “3D MoGe Panorama to Mesh” in Template Library

Download Sample Image

Get the example input image for this workflow
panorama to mesh preview

​
3.1 Steps to Run

  1. Ensure the LoadMoGeModel node has loaded one of the MoGe checkpoints
  2. Load an equirectangular panorama image in the Load Image node
  3. Click Queue or use Ctrl(cmd) + Enter to run

​
Community Resources