Inference

For inference I used ComfyUI.

I uploaded an example workflow here: https://github.com/cseti007/ComfyUI-Workflows/blob/main/wan2_2-lightning1_1-gguf-nfjinx.json

Using the example workflow you can recreate this video: https://civarchive.com/images/93274242

The strength of the LoRA can differ from prompt to prompt. As best practice, I suggest always checking the high model inference and adjusting the high noise LoRA strength or the steps accordingly. Mostly it is optimal when the character features are just beggining to appear in the high model inference, but aren't prominent yet.

• Trigger words: Nfj1nx, blue hair

• Strength: 0.6-1.2

You can find example prompts in the video examples!

Trainig details

Trained only on videos.

HIGH noise LoRA

• dataset: 30 videos 480x270 25,33,65,81 frame videos

• steps: 2130

• LR: 5e-5

• optimizer: AdamW Optimi

• rank: 32

• batch size: 1

• gradient accumulation steps: 1

• min_t = 0.875

• max_t = 1

LOW noise LoRA

• dataset: 42 videos 640x360 25,33,65 frame videos

• steps: 2730

• LR: 5e-5

• optimizer: AdamW Optimi

• rank: 32

• batch size: 1

• gradient accumulation steps: 1

• min_t = 0

• max_t = 0.875

For training I used the diffusion-pipe repo.

Important Notes: This LoRA is created as part of a fan project for research purposes only and is not intended for commercial use. It is based on the movies, which are protected by copyright. Users utilize the model at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The model has been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of the model.