OrbitQuant UNet Loader (INT-Crush):
OrbitQuantLoader loads a diffusion model that has been quantized in the OrbitQuant format for use in a ComfyUI workflow. Quantization represents model weights more compactly; OrbitQuant’s method combines an RPBH rotation with a Lloyd-Max codebook. The node checks the model’s metadata to confirm that it uses the supported OrbitQuant format, then prepares the operations and shared data needed to load it. It also registers adapter synchronization callbacks on the loaded model. You can use it when you have an OrbitQuant model and need to bring it into a workflow without configuring its loading process manually.
OrbitQuant UNet Loader (INT-Crush) Input Parameters:
unet_name
Select the diffusion model file you want to load. The options come from the files ComfyUI lists under diffusion_models; the available names depend on your installation. Your selection determines which file is checked and loaded. That file must contain metadata identifying int_crush.format_version as 4 and int_crush.orbitquant as true. This is a file-selection option, so numeric minimum and maximum values do not apply. The source does not specify a default selection.
OrbitQuant UNet Loader (INT-Crush) Output Parameters:
model
The loaded MODEL object. Connect it to a downstream node that accepts a diffusion model. You receive this output only if the selected file is found, passes the OrbitQuant metadata check, and loads successfully; it is not an image or a model filename.
OrbitQuant UNet Loader (INT-Crush) Usage Tips:
- Choose a file that was exported in OrbitQuant format version 4. The node checks the format metadata and will reject other INT-Crush formats.
- If your model is an INT-Crush INT4 or INT8 file rather than an OrbitQuant file, use
SimpleINT4UNetLoaderorSimpleINT8UNetLoader, respectively, as directed by the node’s validation error. - Select the model from the
unet_namedropdown, then connectmodelto a node that accepts aMODELinput. There are no exposed rotation, codebook, or performance settings to tune in this node. - If a metadata warning appears, check the model file before relying on the result. The loader falls back to defaults when metadata cannot be read, but the OrbitQuant validation may then prevent loading.
OrbitQuant UNet Loader (INT-Crush) Common Errors and Solutions:
Model not found: <unet_name>
- Explanation: ComfyUI could not resolve the selected filename in its
diffusion_modelslocations. - Solution: Check that the file is still present in a configured diffusion-model folder, refresh ComfyUI’s model list if needed, and select the file again.
Not an OrbitQuant model (expected format_version=4 and int_crush.orbitquant=true). Use SimpleINT4UNetLoader or SimpleINT8UNetLoader for other INT-Crush formats.
- Explanation: The file’s metadata does not identify it as the OrbitQuant format this node accepts. Missing or unreadable metadata can also lead to this check failing.
- Solution: Confirm that you selected an OrbitQuant format-version-4 file. For another INT-Crush format, choose the appropriate loader named in the error.
[INT-Crush] Failed to load metadata from '<path>': <error>. Falling back to the node's defaults, which may produce incorrect output if the model was quantized with non-default settings.
- Explanation: This is a warning emitted when the metadata read raises an error. This node subsequently requires OrbitQuant metadata, so loading may also stop with the “Not an OrbitQuant model” error.
- Solution: Check that the selected file is intact and is the intended OrbitQuant model. Review the path and underlying error shown in the warning, then try a valid copy of the file if necessary.
