This week, the race for world models began. Matrix-Game 2.0: An Open-Source, Real-Time, and Streaming Interactive World Model - 25 FPS live video synthesis - Realtime inputs - Trained on ~1200 hours of video data 🤗 available now on Hugging Face

OmniAvatar - Audio-driven avatars with body language

this isn't a 3D model. let's fix that how to create *game-ready* 3D Assets with AI and Blender

The future of Blender is really exciting (using Hugging Face meshgen plugin) - "create a ground plane and randomly scatter around some placeholder bushes" - "add a big oak tree in the middle of the scene" It's a great example of AI as a tool to make 3D artists' lives easier, not to replace them

🚀 Introducing meshgen v0.7 Control Blender with AI Agents - entirely locally ☁️ Like Blender-MCP entirely within Blender 🤗 Supports both local models and remote APIs more details below 👇

Creating a cube with AI in Blender it's kind of like creating a cube without AI, but 100x slower 🎉 but the AI is running completely locally within Blender. major release coming soon

You now have the option to filter 3D Arena to open source models.

Impressive new Image-to-3D model from Tencent! Here's how the topology (left) compares to the open source state-of-the-art (right) (using the `Simplify` option for reduced poly count)

🚀 Announcing meshgen (AI in Blender) Update 0.6 🤗 Now supports remote backend via litellm, e.g. Hugging Face, ollama 🎨 UI/UX overhaul This lays the foundation for Blender agents and more advanced 3D AI models (coming this year)

3D artists have been asking for AI topology tools for a while now really exciting to see generative 3D research that is 1. Iterative 2. Topology-aware definitely a step in the right direction! link to paper below

meshgen (AI in Blender) update 🥳 meshgen now uses the new blender extension system, massively simplifying dependency management this lays the foundation for a lot more to come soon link below

🧊 New dataset of 50k+ low poly obj meshes 🎢 objaverse subsampled to <500 poly models, converted to untextured objs 🔧 suitable for training autoregressive transformer-based 3D models, which have limited context length, such as LLaMA-Mesh

📢 New #1 in Generative 3D CSM/Cube from Common Sense Machines is now the top ranked image-to-3d model

Hunyuan3D-2 has been added to the 3D Arena leaderboard! 🤗 It looks impressive and is definitely a leader, but may still be outperformed by TRELLIS, which is smaller and produces less dense topology

WebGPU Gaussian Splatting VFX 🪄 Unity 6 VFX Graph with WebGPU backend for gaussian splatting with interactive VFX

experimenting with TRELLIS 3DGS + SplatVFX in Unity what kind of game could be made with this?

🟦 New Image-to-3D model from Stability AI Stable Point Aware 3D (SPAR3D) here's how it looks, with TRELLIS for comparison

Here's a recap of what happened in AI 3D in 2024

meshgen v0.3.1 has been released! adds support for: 🦙 Ollama server backend ⚙️ more platforms

TRELLIS is now the highest ranked open-source model on the 3D Arena Leaderboard, surpassing InstantMesh

🟦 New open-source Image-to-3D model from Microsoft TRELLIS: Structured 3D Latents for Scalable and Versatile 3D Generation it's really good! the topology isn't clean, but it's a very very good 3D reference

meshgen 0.3 now uses llama-cpp and quantized LLaMA-Mesh, dramatically improving performance and memory requirements 🔥 loads 10x faster 🚄 runs 5x faster running real-time on RTX 4090 in the video

Generate meshes with AI locally in Blender 📢 New open-source release meshgen, a local blender integration of LLaMa-Mesh, is open source and available now 🤗

Minecraft AI

LLaMa-Mesh running locally in Blender official @huggingface.bsky.social release coming soon 🤗