Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-GNM

ComfyUI Extension: ComfyUI-GNM

Repo Name

ComfyUI-GNM

Author
soylab-edu (Account age: 149 days)
Nodes
View all nodes(1)
Latest Updated
2026-07-16
Github Stars
0.01K

How to Install ComfyUI-GNM

Install this extension via the ComfyUI Manager by searching for ComfyUI-GNM
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-GNM in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-GNM Description

ComfyUI-GNM creates and poses customizable 3D heads in ComfyUI. Adjust identity, expression, and camera view interactively, then render matching color images, normal maps, depth maps, and masks.

ComfyUI-GNM Introduction

GNM Head Viewport outputs: color image, normal map, depth map, and mask

ComfyUI-GNM lets you create and pose a 3D head inside ComfyUI. You can explore different facial identities and expressions, turn the head or camera, and produce four matching outputs:

  • Color image: A rendered view of the head.
  • Normal map: An image describing which direction each visible surface faces.
  • Depth map: An image describing how far visible surfaces are from the camera.
  • Mask: An image separating the head from its background.

The normal and depth maps can be connected to compatible ControlNet workflows. This gives you a way to guide a generated image toward a particular face shape, expression, or viewing angle without relying on a text prompt alone.

The extension uses Google GNM, short for Generative aNthropometric Model, and includes its GNM Head v3 model data. You do not need to download that model separately. For hands-on work, the GNM Head Viewport nodes show an interactive preview; the preview and queued output use the same renderer, so the composition you see is the one you render.

How ComfyUI-GNM Works

Think of GNM as a digital head sculpture controlled by sets of numbers. One set shapes the identity, such as the head and eye structure. Another shapes the expression, such as movements around the eyes and mouth. ComfyUI-GNM uses those numbers to build a head, pose it, and render the four outputs together.

You can create those numbers in two ways:

  1. Sample a starting point. A seed generates a repeatable set of identity or expression values. Changing the seed lets you explore variations; keeping it lets you return to one you like.
  2. Shape specific components. The Advanced and Pro viewport nodes add sliders for individual parts of the model. These slider values are added to the sampled values, allowing you to refine a starting face.

You can also change what the camera sees. Head rotation turns the head itself, while camera azimuth and elevation move the viewpoint around it. For example, turning the head to the side and orbiting the camera to the side are different ways to arrive at a profile-like composition.

The viewport responds as you work, without requiring a Queue run for every adjustment. During interaction, it uses a fast 256 px preview and then refines to 384 px when you stop. Queued output uses the node’s full width and height settings, not the preview resolution.

ComfyUI-GNM Features

Choose a viewport for your workflow

All three viewport nodes provide an interactive preview, mouse-drag camera orbit, wheel zoom, and the same four outputs: image, normal_map, depth_map, and mask. Their main difference is how much direct control they give you over the face.

NodeWhat you controlBest suited to
GNM Head ViewportIdentity and expression sampling, including category, seed, and strength; head and eye pose; camera and render settingsExploring faces and compositions quickly
GNM Head Viewport (Advanced)Sampling plus selected identity and expression component sliders; separate neck and head rotation; gaze, eye convergence, and XYZ positionSculpting prominent features by hand
GNM Head Viewport (Pro Node)Everything in Advanced, with all available identity and expression components in grouped, scrollable controlsFine adjustments to specific features, including teeth or details around the eyes

In the standard viewport, head yaw, pitch, and roll are distributed between neck and head automatically. Advanced and Pro let you rotate the neck and head separately. They also provide vergence, which adjusts how the two eyes converge, and translate XYZ, which moves the head’s position within the frame.

Shape a face with sampling and sliders

The Advanced node exposes 10 head and 3 eye identity components, along with selected expression controls for the eyes, lower face, tongue, and pupils. The Pro Node exposes all 253 identity components (170 head, 3 eye, and 80 teeth) and 383 expression components, organized by region.

Advanced and Pro start with identity_strength and expression_strength at 0. This means you begin with a template, or average, face and can shape it using component sliders. Raise a strength value to bring a seed-sampled identity or expression into the starting point, then use the sliders for adjustments. A group’s reset button clears only that group’s component adjustments. The slider values are saved with the workflow and restored when you reopen it.

As a concrete example from the GNM demonstration, you can experiment with an open-mouth shape in Expression → lower face using e200 = -2.6, e201 = 1.0, and e202 = -2.1.

Control the pose and render

The rendering controls affect either the head, the eyes, or the image you receive:

SettingEffect
head_yaw, head_pitch, head_rollRotate the head pose, in degrees
eyes_yaw, eyes_pitchChange eye direction
camera_azimuth, camera_elevationOrbit the camera; 0/0 is a front view
zoomChange how much of the frame the head fills
fovChange the perspective of the view
surfaceChoose plain for a solid-colored material or edgeflow_texture for the GNM default texture
mesh_partsChoose face_default for the default visible parts, excluding the outer eye shell, or all for all mesh parts

Drag inside a viewport preview to orbit the camera, or use the mouse wheel to zoom. Those changes are reflected in the camera controls and carried into the queued render.

Use separate nodes when you need a modular workflow

You do not have to use an all-in-one viewport. ComfyUI-GNM also provides:

  • GNM Model Loader: Loads GNM Head v3 and caches it after the first load.
  • GNM Random Params: Samples identity and expression vectors from a standard normal distribution.
  • GNM Identity Sampler: Samples an identity conditioned on one of two gender categories and four race categories.
  • GNM Expression Sampler: Samples from 20 expression classes, including examples such as happy, surprise, and wink.
  • GNM Head Render: Takes model and parameter inputs, poses the head, and produces the color image, normal map, depth map, and mask.

The simplest way to begin is to add GNM Head Viewport and connect its image output to a Preview Image or Save Image node. To build a workflow with separate parameter nodes, use the Model Loader and a parameter sampler with GNM Head Render.

ComfyUI-GNM Models

ComfyUI-GNM uses GNM Head v3. The extension includes the model data and a NumPy-based implementation of the model core; there is no separate model download. The supplied information does not describe alternative GNM model versions within this extension.

The names Head Viewport, Advanced, and Pro Node refer to different interfaces for controlling the same head model, not different models. Choose among them based on whether you want quick sampling, a manageable set of sculpting sliders, or access to every component.

Troubleshooting ComfyUI-GNM

The final image looks less detailed than expected

  1. Check the node’s width and height values.
  2. Run the workflow through Queue and inspect the resulting output.
  3. Remember that the interactive preview intentionally uses a lower resolution for responsiveness; it does not set the resolution of the queued image.

The Advanced or Pro face looks like an average face

  1. Check identity_strength and expression_strength. Both default to 0 in these nodes.
  2. If you want a seed-generated starting face, increase the relevant strength.
  3. Adjust component sliders after sampling to refine the result. At zero strength, the sliders instead shape the template face directly.

A viewport adjustment did not appear in a saved image

  1. Check that you are viewing the output from the most recent Queue run.
  2. Run the workflow again after making the adjustment.
  3. Confirm that the node’s output is connected to the Preview Image or Save Image node you are inspecting.

Rendering fails on a headless Linux server

The author notes that a headless Linux environment may require PYOPENGL_PLATFORM=egl for rendering. This is an environment setting rather than an artistic control; ask the person maintaining your ComfyUI server to check it if rendering fails there.

Frequently asked questions

Do I need TensorFlow to use the gender, race, or expression samplers?
No. The extension runs their decoder with NumPy and reads the included weights without requiring TensorFlow.

Will the preview match my queued composition?
Yes. The built-in preview and final output use the same renderer and camera composition. The preview is lower resolution while you interact; the queued render uses your specified output size.

Can I use the depth and normal outputs for image generation?
Yes. Connect them to compatible depth or normal ControlNet conditioning workflows to guide a generated image.

Will my component-slider changes survive when I save a workflow?
Yes. Advanced and Pro store those values in the node’s workflow data, so they are restored when the workflow is loaded.

Learn More about ComfyUI-GNM

ComfyUI-GNM Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-GNM detailed guide | ComfyUI