MiniMax H3 LoRA Recreates Millie Bobby Brown

This is less a new video model than a sharp demonstration of how cheaply celebrity identity can be packaged into an open video stack.

• –A small adapter makes character personalization easy to download, share, and run alongside a much larger base model.
• –Text-triggered identity consistency goes beyond one-off face swapping, though robustness across poses, lighting, and scenes remains unproven.
• –MiniMax H3’s multimodal architecture supports video generation with native audio, making identity LoRAs more useful for complete synthetic clips. [MiniMax H3 documentation](https://github.com/MiniMax-AI/MiniMax-H3)
• –Apache 2.0 licensing covers the repository, not necessarily consent, publicity rights, or training-data provenance for a real person’s likeness.
• –The demo is compelling, but the sparse model card makes independent quality and safety assessment difficult.