ComfyUI-TRELLIS2_Motion

MergeMeshSequences

Merge multiple mesh sequences into one. Useful for combining sequences processed in batches.

TRELLIS2/Video · 2 inputs · 0 parameters · 1 output

MeshRenderer

Render a 3D mesh from a specified camera viewpoint. Supports multiple rendering modes (PBR, wireframe, cel shading, etc.) and lighting presets.

TRELLIS2/Render · 3 inputs · 5 parameters · 1 output

MeshSequenceRenderer

Render a sequence of meshes from a consistent camera angle. Designed to work with VideoToMeshes output, rendering each frame's mesh from the same camera angle for video output.

TRELLIS2/Render · 3 inputs · 7 parameters · 1 output

MultiAngleCamera

Generate multiple camera angles for batch rendering. Creates a grid of camera positions for multi-view rendering.

TRELLIS2/Camera · 0 inputs · 5 parameters · 1 output

OnionSkinning

Add motion ghosting effect to rendered video frames. Creates a visual trail showing previous (and optionally future) frames as transparent overlays, commonly used in animation to visualize motion. Parameters allow control over: - Number of ghost frames - Frame interval between ghosts (every 5 frames, 10 frames, etc.) - Ghost opacity and falloff - Color tinting for past/future ghosts

TRELLIS2/Effects · 3 inputs · 7 parameters · 1 output

RenderingEffects

Apply post-processing visual effects to rendered images. Includes effects like: - Bloom (glow on bright areas) - Depth of field (blur based on depth) - Motion blur - Cel shading quantization - Outline/edge detection

TRELLIS2/Effects · 2 inputs · 5 parameters · 1 output

Trellis2Predict

Convert a single image to a 3D mesh with PBR materials using TRELLIS2. TRELLIS2 (4B parameter model) processes a single image and produces a high-quality 3D mesh with PBR materials (albedo, roughness, metallic). Note: This node requires the TRELLIS2 model and CUDA GPU. First run will download the model (~8GB).

TRELLIS2/Core · 1 input · 5 parameters · 1 output

Trellis2VideoAngleShift

All-in-one TRELLIS2 video angle shift node. This convenience node combines the entire pipeline: 1. Extract video frames 2. Process each frame through TRELLIS2 to create 3D mesh 3. Render all frames from a shifted camera angle 4. Optionally apply onion skinning effect 5. Optionally interpolate skipped frames 6. Return the shifted video For more control, use the individual nodes (VideoToMeshes, CameraTrajectory, MeshSequenceRenderer) instead. Requirements: - TRELLIS2 (pip install git+https://github.com/microsoft/TRELLIS.git) - nvdiffrast (pip install git+https://github.com/NVlabs/nvdiffrast.git) - CUDA GPU

TRELLIS2 · 2 inputs · 14 parameters · 1 output

VideoToMeshes

Batch process video frames through TRELLIS2 to create mesh sequence. Takes a batch of video frames and processes each one (with optional frame skipping) through TRELLIS2 to produce a sequence of 3D meshes. Note: This is compute-intensive. Processing time scales with frame count. Use frame_skip to reduce processing time for long videos.

TRELLIS2/Video · 1 input · 5 parameters · 1 output