In the video
Motion authoring
Loom link not set yet — paste the share URL into LOOM_URL at the top of this file.
- The Loom walks through a selection of tools that give highly fine-grained control over generating motion on a 3D stage.
- These tools are built to drop into the workflows game developers already use (Maya, Blender, Unreal, Houdini, with one-click export in both directions) and collapse several work-hours into a couple of minutes.
- Just as importantly, every one of them is also a tool for our orchestrator-AI, which is what lets an animator direct the stage through text or voice at whatever level of control they want.
- Not shown: mocap cleanup, which takes raw sensor data or ordinary video and returns clean, rig-agnostic motion with correct foot and hand contacts against arbitrary ground topologies and props.
Roadmap
- Next up are self-correcting motion from AI-generated video (the model detects and fixes physically implausible frames on its own), quadruped motion, and photorealistic skinning.
- Phase 2 is owning more of the 3D stage and driving full text/voice direction through the orchestrator-AI.
- Phase 3 moves the representation from twenty-year-old polygon meshes to Gaussian splats for photorealism, which opens the door beyond games into film, episodic and advertising.
Our vision
- Every creator is about to become a 3D creator, whether or not they ever open a 3D tool. The best ones are already calling this their last year shooting in 2D.
- The reason is simple: anything captured in 2D can now be reconstructed in 3D, and once it lives in 3D it can be used. Move the camera, relight the scene, redirect the performance, build the set once and shoot inside it forever. Every angle and every cut stay consistent.
- What's missing is the layer in between: a representation both a human and an AI can read and edit, and a model that can iterate, compose, direct and correct on that stage. That is what Genga Machines is building.
- We start with the hardest, most expensive piece: 3D motion. The model that honors frame-level constraints for a AAA animator is the same model that lets a solo creator direct a scene by voice.
- The rest of the stack is arriving on its own. Asset generation is production-grade (Meshy, Tripo), and frontier LLMs already build full Blender scenes through MCP. Today's 3D tools were built for specialists, and they will look ancient very soon.
- The creative stack for the next generation of visual content is one where every creator works in 3D and never has to think about it.