ComfyUI Node
Gaussians From Point Cloud
A ComfyUI node in GaussianPack with 7 inputs and 1 output.
Gaussians From Point Cloud
- ply_path
◄ply_path—►
◄output_filenamegaussians_from_pcd►
◄output_dir►
◄init_scale1.00►
◄init_opacity0.100►
◄knn_k3►
◄subsample_max0►
CategoryGaussianPack
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| ply_path | STRING | Path to an input PLY. Can be a plain point cloud (x/y/z + optional r/g/b) or an existing 3DGS PLY (in which case all fields are read; missing ones get sensible defaults). knn-derived scales OVERRIDE whatever scales the input had. | |
| output_filename | STRING | gaussians_from_pcd | Basename for the output PLY (no extension). |
| output_diropt | STRING | Directory to write the output PLY. Leave blank to use ComfyUI's output folder. | |
| init_scaleopt | FLOAT | 1.000.01–10 | Multiplier on the knn distance when computing initial scales. 1.0 = exactly the mean distance to k nearest neighbors. <1 = smaller gaussians (sharper but possibly under-covered). >1 = larger (smoother but blurrier). |
| init_opacityopt | FLOAT | 0.1000.001–0.999 | Sigmoid-space initial alpha per gaussian. The node applies logit(init_opacity) and writes that to `opacity`. 0.1 matches upstream 3DGS sfm-init. |
| knn_kopt | INT | 31–10 | k for the knn lookup. 3 matches upstream 3DGS (mean distance to the 3 nearest neighbors after skipping self). Larger k = smoother scale field. |
| subsample_maxopt | INT | 00–50000000 | Optional random cap on the gaussian count. 0 = no cap (write every input point). Set a positive value (e.g. 2_000_000) if a downstream node OOMs at huge counts. Deterministic seed = N_in. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| ply_path | STRING | — |