This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHubThe VAEDecodeHunyuan3D node converts latent representations into 3D voxel data using a VAE decoder. It processes the latent samples through the VAE model with configurable chunking and resolution settings to generate volumetric data suitable for 3D applications.
Inputs
| Parameter | Data Type | Required | Range | Description |
|---|---|---|---|---|
samples | LATENT | Yes | - | The latent representation to be decoded into 3D voxel data |
vae | VAE | Yes | - | The VAE model used for decoding the latent samples |
num_chunks | INT | Yes | 1000-500000 | The number of chunks to split the processing into for memory management (default: 8000) |
octree_resolution | INT | Yes | 16-512 | The resolution of the octree structure used for 3D voxel generation (default: 256) |
Outputs
| Output Name | Data Type | Description |
|---|---|---|
voxels | VOXEL | The generated 3D voxel data from the decoded latent representation |