Nodes/ComfyUI-MeshFlow/MeshFlow Remesh
ComfyUI Node

MeshFlow Remesh

A ComfyUI node in MeshFlow with 15 inputs and 1 output.

By Rizzlord·Created 2 months ago·Updated about a month ago· 12
MeshFlow Remesh
  • trimesh
  • reference_image
  • trimesh
model_namemeshflow
steps28
samplerheun
guidance_scale2.5
seed42
base_num_verts4096
points4096
image_size512
devicecuda
dtypefp16
compilefalse
use_rmbgfalse
fill_holesfalse
CategoryMeshFlow

Inputs (15)

NameTypeDefaultDescription
trimeshTRIMESHThe input 3D model to be remeshed.
model_nameCOMBOmeshflowSelect the MeshFlow model to use. 'meshflow' is the standard model. 'meshflow_w_num_verts_control' allows dynamic control of the generated mesh resolution.
stepsINT281–1000Number of diffusion sampling steps. Higher values can increase detail but take longer.
samplerCOMBOheunODE solver for flow-matching sampling. Higher-order solvers (heun, rk4) produce more accurate trajectories; heun uses 2 model evals per step, rk4 uses 4.
guidance_scaleFLOAT2.50–100Classifier-Free Guidance (CFG) scale for visual conditioning. Only effective when reference_image is connected.
seedINT420–18446744073709550000Random seed for sampling latents.
base_num_vertsCOMBO4096The base resolution/point count of the loaded model checkpoint (sequence length).
pointsCOMBO4096Target resolution (points/vertices) of the generated output mesh.
image_sizeCOMBO512Resolution to resize and center crop the reference image to before processing.
deviceCOMBOcudaComputation device to run the model on (cuda or cpu).
dtypeCOMBOfp16Precision model dtype (fp16, bf16, or fp32).
compileBOOLEANfalseWhether to use torch.compile on CUDA for faster inference.
use_rmbgBOOLEANfalseEnable automatic background removal and foreground cropping for the reference image.
fill_holesBOOLEANfalseAutomatically closes and triangulates boundary loops in the final mesh topology.
reference_imageoptIMAGEOptional reference image for image-conditioned generation.

Outputs (1)

NameTypeDescription
trimeshTRIMESH