Perflow-Shuai/WAN-Fun-LoRA-5B-Control-TrainCFG4-4Step-CFG1-step1600
Video-to-Video • Updated • 26
Downstream Lora
LoRAs trained on downstream video models. Each release states its training checkpoint, inference steps and CFG settings.
Note WAN Fun LoRA: training teacher CFG=4 (RGS=3), training step1600; 4 inference steps, CFG=1 (CFG-free), rank64.
Perflow-Shuai/SCOPE-Wan2.2-5B-NonAR-DMD-4Step-LoRA-r32-iter2000
Image-to-Video • Updated • 12Note SCOPE-specific non-AR DMD LoRA trained on zizhaotong/SCOPE with image, text and 10-DoF actions. Checkpoint step2000; rank/alpha 32/32; 4-step FlowUniPC, shift5; released inference default CFG=5. Generator-only adapter.