This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHubThe LotusConditioning node provides pre-computed conditioning embeddings for the Lotus model. It uses a frozen encoder with null conditioning and returns hardcoded prompt embeddings to achieve parity with the reference implementation without requiring inference or loading large tensor files. This node outputs a fixed conditioning tensor that can be used directly in the generation pipeline.
Inputs
| Parameter | Data Type | Required | Range | Description |
|---|---|---|---|---|
| No inputs | - | - | - | This node does not accept any input parameters. |
Outputs
| Output Name | Data Type | Description |
|---|---|---|
conditioning | CONDITIONING | The pre-computed conditioning embeddings for the Lotus model, containing fixed prompt embeddings and an empty dictionary. |