For T2V:
Training set:
23 video clips of 3 seconds each, mainly 3D animation
Recommended settings:
High Noise: 10 steps, cfg 3.5, shift 6, euler simple.
Low Noise: 4–5 steps, cfg 1, shift 6, lcm simple. Use together with WAN2.2 Lightning 4 steps lora. you can download here: https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Wan22-Lightning
Wan2.2-Lightning_T2V-v1.1-A14B-4steps-lora_LOW_fp16.safetensors
Style:
You can try controlling the style through prompts, such as real-life video or 3D CG art or 2D animation
Since most of the dataset consists of 3D animation, this LoRA often fails to generate real human videos .You can try specifying the character’s ethnicity in the prompt or describing it as a real-life cosplay video.
Description
Training set:
23 video clips of 3 seconds each, mainly 3D animation
use 5070ti for trainning