lucataco / hunyuanvideo-lora-trainer

Fine-tune HunyuanVideo via a-r-r-o-w/finetrainers (Work In Progress)

  • Public
  • 53 runs
  • H100
  • GitHub
  • Weights
  • Paper
  • License

Input

*file

ZIP file containing video dataset

string
Shift + Return to add a new line

Trigger word

Default: "afkx"

integer
(minimum: 10, maximum: 4000)

Number of training steps

Default: 500

integer
(minimum: 16, maximum: 128)

LoRA rank

Default: 128

integer
(minimum: 1, maximum: 4)

Batch size

Default: 1

integer

Gradient accumulation steps

Default: 1

integer

Random seed

Default: 0

string
Shift + Return to add a new line

Hugging Face model path to upload trained LoRA

secret

A secret has its value redacted after being sent to the model.

Hugging Face token for model upload

Output

Generated in

Run time and cost

This model runs on Nvidia H100 GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

About

A Cog implementation of a-r-r-o-w/finetrainers finetune scripts for training video LoRAs

How to use

Upload your training data as a zip file and hit Run to train. Your zip file should contain:

  • prompts.txt file which should contain prompts separated by lines
  • videos.txt file should contain the video file paths, separated by lines
  • mp4 video files at any resolution as long as it is divisible by 32. (For example, 720 * 480, 1920 * 1020, etc) And supported frame counts must be 4 * k or 4 * k + 1 (example: 16, 32, 49, 81). To help with video splitting and resizing see this model: lucataco/video-split

Run your LoRA

After training your LoRA use the following inference model: [lucataco/hunyuanvideo-community-lora]https://replicate.com/lucataco/hunyuanvideo-community-lora)