← News
Realtime AI video games, Moltbook, agent swarms, AI Earth, new AI video models: AI NEWS
This episode is a “what shipped this week?” roundup across AI video, agent tooling, and media models. If you’re skimming, pick one item below and start from the linked demo/paper.
- Nvidia Earth-2 open models: climate / geospatial modeling work that’s increasingly “model zoo”-like.
- Gemini “agentic vision”: Google’s latest framing for tool-using vision models.
- HunyuanImage 3.0 Instruct: a newer instruction-tuned image model worth benchmarking.
- Moltbook: a UI automation workflow tool that fits the “agents controlling apps” trend.
- Project Genie: DeepMind’s ongoing work on world models / game-like simulators.
- Lucy 2: realtime video generation positioned as game-like interactivity.
- TeleStyle: style transfer work for video/telepresence workflows.
- Qwen Image 2512 Turbo LoRA (2 steps): a “fast path” fine-tune for Qwen Image.
- Qwen3 ASR (1.7B): a larger ASR model release from the Qwen line.
- MiniMax Music 2.5: music generation updates from MiniMax.