AI-powered 3D suite built on Blender, with an integrated agent, moodboard, generation tools and layer-based painting

Mixar

Mixar is an open‑source desktop application that extends Blender 5.2 with AI‑driven workflows for 3D creation. It ships the full Blender feature set and adds:

AI chat agent (Mixie) – you can type natural‑language commands to model, texture, fix UVs, set up materials, etc.
Layer‑based texture painting – Photoshop‑style stacked layers, masks, procedural materials, UDIM support and asset export.
AI‑assisted 3D generation – text‑to‑3D and image‑to‑3D mesh generation (via Hunyuan or other models), with automatic retopology and UV unwrap.
Moodboard‑driven scene creation – drop reference images and let the AI build a full scene.
Asset search – neural‑embedding search across your local library.
Bring‑your‑own‑key (BYOK) – optionally route the chat agent through your own OpenAI/Anthropic API keys instead of Mixar’s hosted credits.
Native Mixar editor spaces – dedicated panels for Layers, Properties, Assets, and Chat inside the Blender UI.

The AI features talk to Mixar’s closed‑source backend (requires a free Mixar account), but all non‑AI parts (sculpting, animation, rendering, the layered paint module, file I/O) work offline.

Getting started / building from source

Clone the repo recursively (it contains the Blender source as a submodule):

git clone --recursive https://github.com/Mixar-AI/mixar-texture-painting.git
cd mixar-texture-painting

Initialize submodules and LFS assets:
```
make init
```
Configure the environment (copy the example and edit any backend URLs you need):
```
cp .env.example .env
$EDITOR .env
```
Build the whole application (overlay + CMake + compile):
```
make build   # defaults to Prod (release) build
```
The resulting binary is placed under build/<env>/bin/ (e.g., build/Prod/bin/mixar).

Run it:

make run          # launches the Dev build
make run Prod     # launches the production build

Platform notes

Linux – install the long list of development packages shown in the README, use GCC 14/G++ 14, and set MIXAR_CUDA=0 if you don’t have CUDA/OptiX.
macOS – requires CMake (Homebrew or the app bundle) and the standard Xcode toolchain.
Windows – needs Visual Studio 2022 17.14+ with the C++ workload; the build script handles the overlay and CMake configuration.

Quick test

python -m pip install pytest Pillow numpy
python -m pytest -q   # runs the pure‑Python tests that stub bpy

A smoke‑test (tests/qa/linux_build_smoke.py) can be executed without a backend account to verify that the app starts, loads a scene, renders a small CPU Cycles pass, and saves the result.

License

Mixar‑original code: GPL‑3.0‑or‑later.
Modified Blender files: GPL‑2.0‑or‑later (inherited from upstream).
Parts derived from the ucupaint project: GPL‑3.0‑or‑later with attribution.
Brand assets have a separate non‑GPL brand license.

All files carry SPDX headers and the repository follows the REUSE specification.

Project status

Version 2.0.0 – first public source release, built on Blender 5.2.
The desktop app source is fully available; the AI backend remains closed‑source.
Contributions are not yet open (CLA process pending), but bug reports and build‑help questions are welcomed via GitHub Issues or the Discord community.

Where to get help / community

Discord: https://discord.gg/YVqvkQx8rX (fastest place for build help and usage questions)
GitHub Issues: for reproducible bugs, build problems, licensing questions.
Hosted service: sign‑in at https://mixar.app for AI credits, account management, and official support.

In short: Mixar gives artists a familiar Blender environment enriched with AI chat‑driven modeling, texture‑painting layers, and generative 3D capabilities, while keeping the entire Blender workflow intact. The source repo contains everything needed to compile the desktop client, but the AI service itself stays proprietary.

Mixar-AI/mixar-app