Open-source realism LoRA for MiniMax H3 makes generated humans look real

Affectionate-Map1163 · reddit · 2026-08-11

A developer has trained and open-sourced a new LoRA specifically designed for the MiniMax H3 video model to enhance realism.

This fine-tuned weight aims to solve the uncanny valley effect in AI video generation, making generated human subjects look photorealistic. The author included the weights inside the post for the community to download and test.

Related event: fal Releases LoRA Trainer and Realism People Model for MiniMax H3(5 posts)→

Original post →

More from Multimodal

Multimodal channel →