Wan 2.2 GGUF Lightx2v

I can't remember where I found this workflow originally, but I have modified it to use multiple LoRAs and to make a couple of user interface tweaks. If you recognize this let me know so I can give credit to the original version.

Local I2V generation with Wan 2.2 GGUF models, using lightx2v LoRAs to significantly reduce the number of steps required. I recommend doing some research to find the quantized GGUF model that works best with your GPU and VRAM.

I tend to use 7 steps, which is not bad for a 5-second clip, but the image quality starts to degrade noticeably if you intend to use the last frame to generate additional clips. 8+ steps is better if you will be generating multiple clips.

Pro tip: ComfyUI ignores the negative prompt at a CFG value of 1.0 ... but will process it at 1.1 or higher. If I want to run some quick generations to test a new prompt or LoRA I will set the CFG to 1.0, and if I want a high-quality video I run it at 1.1, which takes maybe 20% longer to generate.