Hugging Face
01
Physical benchmark
PhyGround
An open benchmark for the physical plausibility of text-and-image-to-video generation, with 250 curated prompts, a 13-law physics taxonomy, and the PhyJudge-9B evaluator.
Full-Stack Open AI Ecosystem
Truly Open,
We are building a version of the AI future that is open, efficient, and sovereign. From Edge to Cloud.
Physical Intelligence
World Models & Embodied AI
Moxin extends its open AI ecosystem from digital intelligence into the physical world, exploring systems that simulate, predict, and act. NU World Model & Embodied AI openly releases models, benchmarks, datasets, and embodied-agent research.
02
Video world model
PhyWorld
A physics-aware video-generation world model post-trained with temporal-coherence tuning and preference optimization derived from the PhyGround human-annotation pool.
A shared research loop
Measure physical reality. Train against it.
PhyGround provides the physical benchmark and human preference data; PhyWorld uses the same signal to build a more physically coherent video-generation world model.
Desktop App
Moxin Studio
A native desktop AI app built with pure Rust and Makepad. Run everything locally on your Mac, no Python required.
- 20+ Local Models: LLMs, VLMs, image generation, ASR, TTS, and video — all on-device.
- Model Hub: Discover, download, and run models with one click.
- Voice & Vision: Voice cloning, speech recognition, image generation, and visual understanding.
Voice AI
Moxin Voice
Real-time live translation, TTS, and voice cloning running entirely on-device in pure Rust.
- Zero-Shot Voice Cloning: Create a voice profile from just 5-10 seconds of reference audio.
- Local TTS + ASR: No cloud APIs, no latency. 14+ preset voices with custom training support.
- Native Performance: Built with Rust, Makepad, and GPT-SoVITS v2. Deploying at GOSIM Paris 2026.
Data Sovereignty
Your data never leaves your infrastructure. Run fully private AI models on-premise or in your private cloud.
Extreme Efficiency
Run 70B+ models on consumer hardware. Ominix Edge optimizes inference for up to 30x lower latency.
Full Control
Open source from top to bottom. Modify the model, the agent framework, or the inference engine to fit your needs.