Trellis2TextureStage - ComfyUI Built-in Node Documentation

This node sets up the texture-stage sampling pass for Trellis2 generation. It reads the coordinate layout and the per-voxel shape latent from the incoming shape latent, builds an empty sparse latent with 32 channels at the same coordinate layout, and attaches the required texture-stage metadata to the conditioning.

Inputs

Parameter	Description	Data Type	Required	Range
`positive`	The positive conditioning used for the texture-generation pass. Texture-stage metadata is attached to it.	CONDITIONING	Yes	-
`negative`	The negative conditioning used for the texture-generation pass. Texture-stage metadata is attached to it.	CONDITIONING	Yes	-
`shape_latent`	The latent dict produced by Trellis2ShapeStage or Trellis2UpsampleStage. It must contain `coords` (the coordinate layout, shape [N, 4]) and `samples` (the per-voxel shape latent); `coord_resolution` and `model_frame` are optional.	LATENT	Yes	-

Notes:

shape_latent must be the output of Trellis2ShapeStage or Trellis2UpsampleStage; it provides the coordinate layout and the per-voxel shape latent used by the texture pass.
The coordinate layout is validated: batch ids in the first column of coords must be non-negative and contiguous, and the total row count must match the coordinate counts.
When positive carries a projection feature pack (Pixal3D conditioning) and shape_latent includes coord_resolution, projection features at 1024 texture resolution are computed and attached to the conditioning.
The model frame is read from shape_latent; when absent, it defaults to "y_up" if projection features are present and "z_up" otherwise.

Outputs

Output Name	Description	Data Type
`positive`	The positive conditioning with texture-stage metadata attached (generation mode, coordinates, coordinate counts, shape latent, model frame, and projection features when applicable).	CONDITIONING
`negative`	The negative conditioning with the same texture-stage metadata attached.	CONDITIONING
`latent`	A new empty sparse latent with 32 channels at the same coordinate layout as the incoming shape latent. Its dict includes `samples`, `type` (“trellis2”), `coords`, `coord_counts`, and `model_frame`; `coord_resolution` is included when available.	LATENT

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): ae612021af7c74cd09206d905e7b800fa48367a22daf9b0335b444c854a78b1e

​Inputs

​Outputs

Inputs

Outputs