164 likes
FlashGen LoRA slashes MiniMax video generation to four steps
What is minimax-h3-flashgen-4step?
The Gradio demo runs MiniMax-H3, a 33-billion-parameter model that creates video with native stereo audio, using a FlashGen LoRA that reduces diffusion steps from about twenty to four. That change speeds up inference by roughly five times while keeping the model’s audio-video alignment.
Who it is for
Developers and creators who need fast multimodal generation can test the demo or add the LoRA to their own pipelines for quicker results.
How to try it
Open the Hugging Face Space at https://huggingface.co/spaces/hugging-apps/minimax-h3-flashgen-4step