H3-World: Turning Language Understanding into World Control
retained text
* Language-Native Control: Composes character and camera actions into textual instructions and injects them through MiniMax-H3’s pretrained text pathway. * Temporally Grounded: Assigns one action prompt to each video latent interval, enabling precise control when actions change over time. * Efficient & Generalizable: Uses only 8,000 gameplay samples, 10,000 LoRA steps, and 0.199% trainable parameters to achieve controllable character and camera motion, including unseen action compositions and visual scenarios. ✏️ Paper: [https://huggingface.co/papers/2609.01560](https://huggingface.co/papers/2609.01560) 📄 ArXiv: [https://arxiv.org/abs/2609.01560](https://arxiv.org/abs/2609.01560) 💻 Code: [https://github.com/Danzer1xxxxChan/H3-World](https://github.com/Danzer1xxxxChan/H3-World) 🏠 Project: [https://danzer1xxxxchan.github.io/H3-World/](https://danzer1xxxxchan.github.io/H3-World/) 🤗 Model: [https://huggingface.co/DANNY621/H3-World](https://huggingface.co/DANNY621/H3-World)
document fetched 2026-09-03 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1w5akpy/h3world_turning_language_understanding_into_world