Pixal3DConditioning - ComfyUI Built-in Node Documentation

This node prepares image conditioning for the Trellis2 3D generation pipeline. It extracts visual features from the input image with a DINOv3 vision model at two resolutions, organizes them into per-stage feature maps (optionally enhanced with a NAF model), and combines them with camera data derived from the horizontal field of view. It outputs a positive and a negative conditioning pair, where the negative uses zeroed features for classifier-free guidance.

Inputs

Parameter	Description	Data Type	Required	Range
`clip_vision_model`	DINOv3 ViT-L/16 ClipVision.	CLIP_VISION	Yes	—
`image`	Preprocessed image from ImageCropToMask (pad_factor=1.1 for Pixal3D).	IMAGE	Yes	—
`camera_angle_x`	Horizontal FOV in degrees (display name: fov). Wire a MoGeGeometryToFOV (axis=‘horizontal’, unit=‘degrees’) for a per-image FoV (matches upstream default). Default: 49.13.	FLOAT	Yes	1.0 – 170.0

Note: The camera_angle_x value is converted to radians internally and used to compute the camera distance for the projection transform matrix. When the supplied vision model includes a NAF component, the node additionally produces high-resolution feature maps for the shape and texture stages.

Outputs

Output Name	Description	Data Type
`positive`	Positive conditioning containing the image-derived feature maps and projection data for Trellis2 generation.	CONDITIONING
`negative`	Negative conditioning with zeroed feature tensors, used for classifier-free guidance.	CONDITIONING

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 3eba711620f6c56a21bbf7df89f8d406ce6f90908298b1a295a1dbbddd042472

​Inputs

​Outputs

Inputs

Outputs