Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-Trellis2-MultiViewRefiner > Trellis2 - Mesh Refiner Multi-View

ComfyUI Node: Trellis2 - Mesh Refiner Multi-View

Class Name

Trellis2MeshRefinerMultiView

Category
Trellis2Wrapper
Author
cuzelac (Account age: 5597days)
Extension
ComfyUI-Trellis2-MultiViewRefiner
Latest Updated
2026-04-19
Github Stars
0.0K

How to Install ComfyUI-Trellis2-MultiViewRefiner

Install this extension via the ComfyUI Manager by searching for ComfyUI-Trellis2-MultiViewRefiner
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-Trellis2-MultiViewRefiner in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

Trellis2 - Mesh Refiner Multi-View Description

Refines an existing 3D mesh with Trellis2 using multiple views and shape and texture controls.

Trellis2 - Mesh Refiner Multi-View:

Trellis2MeshRefinerMultiView refines an existing 3D mesh using a Trellis2 pipeline and images of the object from multiple directions. It uses a front view and can incorporate back, left, and right views, giving the refinement process more visual reference than a front image alone. The node provides separate controls for shape and texture sampling, along with controls for resolution, view blending, and the sampling method. Its purpose is to help you produce a refined mesh that better reflects the supplied views; the exact appearance depends on the connected pipeline, starting mesh, images, and settings.

Trellis2 - Mesh Refiner Multi-View Input Parameters:

pipeline

The TRELLIS2PIPELINE connection supplies the pipeline that performs refinement. The node calls that pipeline's multiview refinement method, so it must be a compatible Trellis2 pipeline. There is no numeric range or default; you must connect a pipeline.

trimesh

The TRIMESH connection supplies the existing 3D mesh to refine. This is the starting geometry, not an image from which the node creates a mesh independently. There is no default; connect the mesh you want to improve.

front_image

The required IMAGE connection supplies the front-view visual reference. The node converts it to a PIL image before passing it to the pipeline. Its image converter accepts a height-by-width-by-channels tensor, or a batch containing exactly one such image. There is no default.

seed

Controls the seed passed to the refinement pipeline. Keeping the same seed helps you compare setting changes under the same seed value, although the supplied code does not establish a guarantee of identical results. Minimum: 0; maximum: 2147483647; default: 12345.

resolution

Sets the resolution value passed to multiview refinement. It is a useful setting to test when balancing output detail against available resources, but the supplied code does not define a specific memory or quality relationship. Options: 512, 1024, 1536; default: 1024.

shape_steps

Sets the number of sampling steps for the shape stage. More steps mean that stage performs more sampling work; they are not a guarantee of a better mesh. Minimum: 1; maximum: 100; default: 12.

shape_guidance_strength

Sets the guidance-strength value for shape sampling, affecting how strongly that stage uses its guidance. The exact visual effect is determined by the connected pipeline, so compare results on the same mesh and images when adjusting it. Minimum: 0.00; maximum: 99.99; default: 6.50; step: 0.01.

shape_guidance_rescale

Sets the guidance-rescale value for shape sampling. This is a separate adjustment to shape guidance, not a replacement for shape_guidance_strength; its precise effect is implemented by the pipeline. Minimum: 0.00; maximum: 1.00; default: 0.05; step: 0.01.

shape_rescale_t

Passes a rescale_t value to the shape sampler. It adjusts a sampler-specific aspect of shape generation; the supplied node does not define what a particular value looks like, so use the default as your baseline. Minimum: 0.00; maximum: 9.99; default: 4.00; step: 0.01.

texture_steps

Sets the number of sampling steps for the texture stage. This controls that stage independently of shape_steps. More steps require more sampling work but do not guarantee a visible improvement. Minimum: 1; maximum: 100; default: 12.

texture_guidance_strength

Sets the guidance-strength value for texture sampling. It lets you adjust texture guidance without changing shape guidance; the exact visual effect depends on the pipeline. Minimum: 0.00; maximum: 99.99; default: 3.00; step: 0.01.

texture_guidance_rescale

Sets the guidance-rescale value for texture sampling. It modifies the texture stage's guidance settings separately from shape guidance, with its precise behavior determined by the pipeline. Minimum: 0.00; maximum: 1.00; default: 0.20; step: 0.01.

texture_rescale_t

Passes a rescale_t value to the texture sampler. This is a sampler-specific texture setting whose precise visual effect is not defined in the supplied node code. Minimum: 0.00; maximum: 9.99; default: 3.00; step: 0.01.

max_num_tokens

Sets the maximum-token value passed to the refinement pipeline. The node does not show how the pipeline uses or counts tokens, so do not treat this value as a predictable polygon count. Minimum: 0; maximum: 999999; default: 999999.

generate_texture_slat

Controls whether the pipeline is asked to generate a texture SLAT. It also determines whether this node attempts to build a BVH: when True, it builds one after refinement; when False, it returns None for bvh. Options: True or False; default: True.

downsampling

Selects the downsampling value passed to the refinement pipeline. The supplied code does not specify which internal operation it changes or establish a particular quality-to-speed tradeoff. Options: 16, 32, 64; default: 16.

shape_guidance_interval_start

Sets the start of the guidance interval passed to the shape sampler. Together with shape_guidance_interval_end, it defines the interval during which shape guidance is configured to apply. Minimum: 0.00; maximum: 1.00; default: 0.10; step: 0.01.

shape_guidance_interval_end

Sets the end of the guidance interval passed to the shape sampler. Adjust it alongside the start value when changing the portion of shape sampling covered by guidance. Minimum: 0.00; maximum: 1.00; default: 1.00; step: 0.01.

texture_guidance_interval_start

Sets the start of the guidance interval passed to the texture sampler. It is independent of the shape guidance interval. Minimum: 0.00; maximum: 1.00; default: 0.00; step: 0.01.

texture_guidance_interval_end

Sets the end of the guidance interval passed to the texture sampler. Together with the texture start value, it defines the interval supplied for texture guidance. Minimum: 0.00; maximum: 1.00; default: 0.90; step: 0.01.

use_tiled_decoder

Passes the selected tiled-decoding setting to the refinement pipeline. The supplied node does not describe the decoder's internal behavior or quantify its resource use. Options: True or False; default: True.

front_axis

Tells the refinement pipeline which axis represents the front direction for the supplied views. Choose the option that matches your mesh's orientation; a mismatch can make the image directions inconsistent with the mesh. Options: z or x; default: z.

blend_temperature

Passes a temperature value for multiview blending to the refinement pipeline. It gives you a way to adjust blending, but the supplied code does not establish how higher or lower values change the result. Minimum: 0.1; maximum: 10.0; default: 1.0; step: 0.1.

sampler

Selects the sampling method passed to the refinement pipeline. The node exposes four methods, but the supplied code does not establish which is fastest or best for a given subject. Options: euler, heun, rk4, rk5; default: euler.

back_image

An optional IMAGE connection for the back view. When connected, the node converts it to a PIL image and passes it to multiview refinement; when absent, it passes None. The converter accepts one image, not a batch of multiple images. Default: not connected.

left_image

An optional IMAGE connection for the left view. It supplies another visual reference to the pipeline when connected and is passed as None otherwise. The converter accepts one image, not a batch of multiple images. Default: not connected.

right_image

An optional IMAGE connection for the right view. It supplies another visual reference to the pipeline when connected and is passed as None otherwise. The converter accepts one image, not a batch of multiple images. Default: not connected.

Trellis2 - Mesh Refiner Multi-View Output Parameters:

mesh

The MESHWITHVOXEL output is the refined mesh returned by the pipeline. It is the result to use in a compatible downstream mesh workflow. Despite the output type's name, the supplied node code does not describe its internal voxel representation or guarantee which visual details the refinement will preserve.

bvh

The BVH output is a bounding-volume hierarchy built from the refined mesh's vertices and faces when generate_texture_slat is True. It is a spatial structure for compatible downstream operations, rather than another visible mesh. When generate_texture_slat is False, this output is None; the node's log describes BVH construction as being used for texturing.

Trellis2 - Mesh Refiner Multi-View Usage Tips:

  • Start with a connected pipeline, an existing mesh, and a front image. Add back, left, and right images when you have views that correspond to those directions.
  • Check front_axis before refining a mesh with a different front orientation. Use z or x according to how the connected workflow orients that mesh.
  • Begin with the defaults, including resolution of 1024, 12 shape steps, 12 texture steps, and the euler sampler. Change one setting at a time so you can tell which change affected the result.
  • For an initial run at a lower selectable resolution, try 512. If the result warrants another run and your environment supports it, compare it with 1024 or 1536; the supplied code does not quantify the resource cost of those choices.
  • Keep generate_texture_slat enabled when a downstream step needs bvh. Disable it only when your workflow can accept a None BVH output.
  • Supply each view as a single image. The node's converter explicitly rejects an image tensor whose batch contains more than one image.

Trellis2 - Mesh Refiner Multi-View Common Errors and Solutions:

ComfyUI-Trellis2-MultiViewRefiner is out of date: <pipeline class> is missing expected helpers <helpers>.

  • Explanation: The connected pipeline class lacks helpers this extension expects. The node stops before refinement rather than proceeding with an incompatible pipeline.
  • Solution: Check that your Trellis2 installation and multiview-refiner extension are compatible, then update the extension or restore compatible versions together.

ComfyUI-Trellis2-MultiViewRefiner is out of date: pipeline module <module> is missing <attributes>.

  • Explanation: The pipeline's module lacks components required by the extension.
  • Solution: Check the compatibility of the installed Trellis2 package and multiview-refiner extension. Update or reinstall compatible versions, then restart the environment so the installed modules reload.

'FlowEulerMultiViewGuidanceIntervalSampler' not found in ComfyUI-Trellis2's samplers module.

  • Explanation: The node explicitly checks for this sampler support before refinement and stops if it is missing.
  • Solution: Use a Trellis2 installation that provides the required sampler alongside a compatible version of the multiview-refiner extension.

tensor2pil expects batch of 1, got batch=<number>

  • Explanation: One of the connected image inputs contains a batch of multiple images, but the converter handles only one image per view.
  • Solution: Select one image for each connected front, back, left, or right socket before running the node.

tensor2pil expects (H,W,C) or (1,H,W,C), got shape=<shape>

  • Explanation: A connected image tensor does not have either of the shapes accepted by the node's converter.
  • Solution: Feed a standard single-image IMAGE output into the affected view socket and check upstream nodes that reshape or batch images.

tensor2pil expected torch.Tensor, got <type>

  • Explanation: A connected view supplied something other than the tensor type expected by the image converter.
  • Solution: Connect a compatible IMAGE output rather than passing an image object or other value directly.

CUDA or GPU memory error during refinement or BVH construction

  • Explanation: The node uses CUDA for the refined mesh's vertices and faces and, when texture SLAT generation is enabled, for BVH construction. The exact error text depends on the failing operation.
  • Solution: Confirm that the pipeline runs in a compatible CUDA environment. If memory is the issue, try the selectable 512 resolution or a less demanding surrounding workflow; if your downstream workflow does not need bvh, you can also test with generate_texture_slat set to False.

Missing cumesh or failure while building cuBVH

  • Explanation: With generate_texture_slat set to True, the node imports cumesh and constructs a cuBVH from the refined mesh. An unavailable or incompatible dependency can stop execution at that stage.
  • Solution: Check the cumesh dependency required by your compatible Trellis2 setup. If your workflow does not need the BVH output, set generate_texture_slat to False and expect bvh to be None.

Downstream step cannot use bvh because it is None

  • Explanation: The node deliberately returns None for bvh when generate_texture_slat is False.
  • Solution: Set generate_texture_slat to True when the next node requires a BVH, or use a downstream path that does not require one.

Trellis2 - Mesh Refiner Multi-View Related Nodes

Go back to the extension to check out more related nodes.
ComfyUI-Trellis2-MultiViewRefiner
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Trellis2 - Mesh Refiner Multi-View