You can now fine-tune open-source video models

I think this is legit the best implementation of this I've seen. The cinematography you can get from training on videos instead of photos really improves on the default Hunyuan "slow motion" style imo