LTXV Enhance A Video KJ(LTXVEnhanceAVideoKJ)

This node applies the "Enhance-A-Video" technique from the NUS HPC-AI Lab to improve the quality and temporal consistency of video generation. It modifies the input model's attention layers based on the provided latent and a weight parameter.

Inputs

• model (MODEL): The LTXV model to be enhanced.
• latent (LATENT): The video latent that guides the enhancement.
• weight (FLOAT): Strength of the enhancement effect. Default 4.0, range 0.0 to 100.0, step 0.001. Higher values apply stronger modifications.

Outputs

• MODEL: The enhanced model with modified attention for better video quality.

Usage Notes

The weight parameter controls the influence of the enhancement. Start with the default value of 4.0 and adjust based on visual results; very high values may introduce artifacts. This node is intended for use with LTXV models and is applied at inference time.