Hey, guys!
I’m tired of opening Blender once every six months, looking at those buttons, and realizing that I’ve forgotten again where the extruding is. I kind of love it. I kind of want to make 3D. But every time it’s like the first time.
In short, I made Mixer — an open‑source 3D editor that works via LLM. You write in text — it builds the model.
⚠️ This is for you if:
- you don’t have access to top‑tier models (they do 3D right in the chat)
- you’re not a professional 3D artist
- you like local LLMs Not for you if: you’re spinning up subdivisions in Blender every day and you’re fine with that.
Under the hood — GLB/GLTF. This is essentially JSON that describes a 3D object. Convenient: the LLM doesn’t need to «draw» — it just writes the markup, and the application turns it into a model. It’s like if you were dictating to a sculptor over the phone, and he was chiseling away.
You can upload references — their geometry will be sent to the LLM along with the prompt. Like «make it like this». If there are many references and the model’s context is small, send simplified geometry (see the settings). It’s like a sketch instead of a detailed drawing: the essence is the same, but it takes up less space.
You can also upload an existing object and switch to edit mode to modify it.
| Situation | Geometry of the active model | Geometry of the selected node | Geometry of the references | Animation/segment |
|---|---|---|---|---|
| Nothing is selected, there is an active model | ❌ does not leave | ❌ | ✅ leaves | ❌ |
| A node is selected | ✅ leaves the entire model | ✅ leaves | ✅ leaves | ❌ |
| A node + a time segment is selected | ✅ leaves the entire model | ✅ leaves | ✅ leaves | ✅ leaves the segment |
| Segment without a knot | ✅ the entire segment is removed | ❌ | ✅ the segment is removed | ✅ the segment is removed |
| Changing the active model | resets the knot and the segment | — | — | — |
Local models — three ways:
Disclaimers: the program weighs ~567 MB. Yes, a lot. Inside, there’s a built‑in llama‑server. If you want to reduce the build size, build it without it. Tauri v2 is lightweight on its own (it’s Rust, after all), it’s the llama‑server that’s the issue.
The project is open, MIT license — fork, break, send a PR.
I’m developing it as I run into limitations myself. So, periodically click «Update» — maybe there’s something new.
Links below. Happy modeling 😉
[GitHub](https://github.com/ArtemPodloboshnikov/Mixer) source code
Builds for [Windows/Linux/Mac](https://github.com/ArtemPodloboshnikov/Mixer/releases)