Nodes/ComfyUI-Hunyuan3d-2-1/Hunyuan 3D 2.1 Use MultiViews
ComfyUI Node

Hunyuan 3D 2.1 Use MultiViews

The bypass — feed the bake node views you already have, without re-running the generator

By visualbruno·Created about a year ago·Updated 5 days ago· 375
Hunyuan 3D 2.1 Use MultiViews
  • trimesh
  • camera_config
  • albedo
  • mr
  • pipeline
  • albedo
  • mr
  • camera_config
view_size512
texture_size1024

Hy3DMultiViewsGenerator is expensive - it runs the paint diffusion model to create views of your mesh. Hy3D21UseMultiViews is the node that says "forget generating, I already have views." It takes a mesh plus existing albedo and MR images and loads them into a paint pipeline, ready to be baked - no diffusion run, no GPU minutes wasted. It's the reload button for the texturing pipeline.

The name is the whole story: you use multi-views instead of making them. Feed it a TRIMESH, a camera_config (matching the views you have), the albedo and mr image sets, plus view_size (default 512) and texture_size (default 1024) to configure the pipeline. Out come the same pipeline, albedo, mr, and camera_config that the generator produces - ready for Hy3DBakeMultiViews.

How it works

It constructs a Hunyuan3DPaintPipeline from the camera config and view/texture sizes, then calls pipeline.load_mesh(trimesh) to bind the geometry. That's it - two steps. The heavy lifting already happened elsewhere (either you ran the generator earlier, or the views came from disk). The node is pure state setup, which is why it's instant.

When you'd actually use this

Three situations. First, you ran Hy3DMultiViewsGenerator, edited the albedo/MR views in an image editor or ran them through an upscaler (the Mesh_Texturing_UpscaleWithAuraSR example does the upscale-between approach), and now want to bake the improved views. Second, you're building a re-bake loop where only the mesh changed but the camera rig and views didn't. Third, you want to test different texture_size/view_size bake settings without regenerating views each time - this node lets you swap sizes cheaply.

Installing it

Standard pack install; it's pure Python state handling, no C++ extensions needed to construct the pipeline (though baking the result still needs the renderer):

cd ComfyUI/custom_nodes
git clone https://github.com/visualbruno/ComfyUI-Hunyuan3d-2-1
cd ComfyUI-Hunyuan3d-2-1
pip install -r requirements.txt

Common issues

The classic mistake is mismatched state: your albedo/MR views came from one camera rig, but the camera_config you feed says different angles. The bake will happily project views onto surfaces they were never rendered for, producing smears. Always keep the original rig that generated the views. Second gotcha: view count must match the number of cameras in the config - six views, six cameras. And if you're trying to load views saved to disk by the WithMetaData flow, use Hy3D21UseMultiViewsFromMetaData instead, which reads the metadata file and figures out the camera config and file names for you.

CategoryHunyuan3D21Wrapper

Inputs (6)

NameTypeDefaultDescription
trimeshTRIMESH
camera_configHY3D21CAMERA
albedoIMAGE
mrIMAGE
view_sizeINT512
texture_sizeINT1024

Outputs (4)

NameTypeDescription
pipelineHY3DPIPELINE
albedoIMAGE
mrIMAGE
camera_configHY3D21CAMERA