ComfyUI-PBRFusion4

COMFYUI-PBRFusion4

ComfyUI custom nodes for generating depth maps and normal maps from images using the PBRFusion4 diffusion model. Designed for PBR texture workflows.

Installation

Clone this repo into your ComfyUI custom_nodes folder:

cd ComfyUI/custom_nodes
git clone https://github.com/Night1099/COMFYUI-PBRFusion4

The model (PBRFusion4.safetensors, ~4.1 GB) is automatically downloaded on first use from HuggingFace and saved to ComfyUI/models/pbrfusion4/. No manual download required.

Nodes

PBRFusion4 Simple Depth & Normal Generator

Generates a depth map and normal map from an input image.

Inputs:

• image - Input image
• intensity_blend - Blend factor for mixing original image intensity into normal map generation (0.0 - 1.0)
• opengl_normal - When enabled, outputs OpenGL-format normals (Y up). When disabled, outputs DirectX-format (Y down)
• optimize - Limit processing resolution to 1024px max side for faster inference, then resize back to original
• bilateral_d / bilateral_sigma_color / bilateral_sigma_space - Bilateral filter settings for depth smoothing

Outputs:

• depth - Raw depth map
• depth_filtered - Bilateral-filtered depth map
• normal - Normal map generated from filtered depth
• intensity - Extracted intensity map from input

Utility Nodes

• Normal Map Flip Y - Convert between OpenGL and DirectX normal map formats
• Black Threshold Filter - Set near-black pixels to pure black (clean up artifacts)
• Smart Upscale Calculator - Calculate target dimensions and scale factor for 1k/2k/4k output
• Clamp Resolution - Cap dimensions to a max size while preserving aspect ratio
• Conditional Upscale - Upscale using a model with a bypass toggle

Requirements

• diffusers
• huggingface-hub
• numpy
• opencv-python-headless
• safetensors
• torch
• transformers

The nodes use OpenCV for image processing without GUI features, so opencv-python-headless is sufficient for ComfyUI installs.

License

This model is licensed under the https://www.apache.org/licenses/LICENSE-2.0.