ComfyUI Merges UltraTex Multi-View Texturing and Mesh Texture Rebaking Tools

A useful AI-generated 3D model needs more than an attractive render from one camera angle. When an asset is rotated, mismatched colors, visible seams and inconsistent materials can become obvious. On November 9, the ComfyUI project merged a substantial code contribution focused on multi-view texturing and baking the result into a mesh's UV texture.

The merged pull request adds support for UltraTex, described in the contribution as a FLUX.2 Klein 9B LoRA approach that prepares multiple views of a target mesh and uses a reference image to guide its appearance. The implementation arranges a set of views into a grid and supports background-token handling. It also introduces nodes for texturing conditioning, camera orbits and rebaking images back onto a mesh.

The new BakeTextureFromViews node projects edited or generated views onto a UV layout and outputs base-color data together with a confidence mask. The proposal includes depth-edge rejection and an option to fill areas that have not been seen directly. These implementation details matter because they address a common production gap: getting an attractive image from a model is different from obtaining a consistent texture that an engine can actually use on a three-dimensional object.

The GitHub pull request shows a merge into ComfyUI's main branch on November 9. That is a verified source-code integration, not proof that every desktop installation or a previously tagged stable package already includes the nodes. The contributor links model weights and an example workflow, but performance requirements and practical quality depend on mesh topology, UVs, input images, compute hardware and the precise installed version.

A game-art team could test the pipeline on a single low-complexity prop, such as a crate or a stone statue. Prepare a reference image, generate or edit the orbit views and bake them to a UV texture. Then inspect the model in Blender or the target game engine under rotating light. Look for seams, missing areas, incorrect roughness and inconsistent detail between faces, not just whether the front view looks impressive.

This is a more directly relevant development than a simple community-node catalog addition: the merge adds mesh-related functionality in the core codebase. It still needs validation in a real pipeline before developers can treat it as a dependable replacement for conventional texture authoring.