Show HN: Blender for AI Agents
www.mixar.appCreating 3D is hard. LLMs seem to be getting better at tool use and spatial understanding. While MCPs have proved to be a good way to use these tools- the current methods have these challenges:
- Access to scene graph and core C modules of Blender
- Lack of parallelism, only way is to run blender headless
- Lack of deterministic and fast verification layer
- Inference stack- only way to use inference is to hook another MCP
We're building Mixar, think Cursor for 3D. One access point to all generative inference, an agent to build scenes/blockouts, do boring stuff like UVs and export standard formats(.glb/gltf/obj/gbx/usd etc.)
Try it: https://mixar.app - first week is free. Would love to get your feedback.
I’m a bit skeptical about LLMs actually being able to perceive 3D context properly. I think a well-designed structured scene graph could be promising.
I have tried a lot of AI 3D stuff but nothing is usable in actual games. Sometimes it makes my character rigs so weird that I am unable to animate it.
true, what we have seen, agent can place the rigg decently well but where it's performs bad is assigning the weight paints. we're working to solve this. more updates soon!
I’m so bullish on the idea that LLMs will solve spatial understanding themselves. World model makers can succ ma d.
Does the agent have any guardrails around scene complexity before it makes changes?
it has routing among tools basis scene complexity- it can user primitves, 3d asset generation models or pre-existing assets from library, basis the scene requirements.