You can now fine-tune open-source video models

A new tool enables users to fine-tune AI video generation models with their own content. Users create a training dataset from video clips, train a custom model, and generate new videos with personalized styles or characters. The process involves using a YouTube URL or video file to create the dataset, then training and running the model on Replicate's platform. Based on open-source video models like Hunyuan-video, the tool captures unique motion styles. Fine-tuning typically takes 5-10 minutes with default settings.