Lightx2v: CFG and Step Distillation LoRAs for Wan2.1
Lightx2v provides CFG and step distillation LoRAs for Wan2.1, enabling faster inference with fewer steps. Available in multiple rank sizes (4 to 256) for both I2V and T2V.
Lightx2v
Wan2.1DistillationLoRAFast InferenceCFG and step distillation LoRAs for Wan2.1, enabling faster inference with fewer steps. Supports multiple rank sizes (4 to 256) for both I2V and T2V pipelines.
| Developer | Lightx2v |
| Architecture | Wan2.1-based distillation LoRAs |
| Model Type | LoRA (distillation) |
| Capabilities | CFG distillation, step distillation, I2V and T2V |
| Rank Sizes | 4, 8, 16, 32, 64, 128, 256 |
Overview
Lightx2v provides CFG and step distillation LoRAs for Wan2.1 that enable significantly faster inference with fewer sampling steps. By distilling the classifier-free guidance (CFG) process and reducing the required number of diffusion steps, these LoRAs can accelerate video generation while maintaining high visual quality.
The LoRAs are available in multiple rank sizes ranging from 4 to 256 for both image-to-video (I2V) and text-to-video (T2V) pipelines, allowing users to choose the optimal balance between speed and fidelity for their hardware capabilities.
ComfyUI Integration
Lightx2v LoRAs are compatible with ComfyUI WanVideo native nodes and Kijai's WanVideoWrapper. Apply the LoRA after loading the base Wan2.1 model to benefit from accelerated inference with reduced step counts.
Model files are available on the Kijai/WanVideo_comfy repository.
Guides and workflows related to this model series.