Developer Tests MiniMax H3 Video LoRA: 4-Step Fast Generation

ostrisai · x · 2026-08-04

Developer ostrisai is training a turbo time LoRA for the MiniMax H3 model. The algorithm is still being adjusted to handle audio (currently broken), and a better, non-slow-motion dataset is being generated.

The shared comparison samples demonstrate the visual output using only 4 steps, highlighting the generation speed boost provided by the LoRA.

Original post →

More from Multimodal

Multimodal channel →