bevy_synth 🕊️🔥🎛️🎹

asset synthesis implemented in bevy and burn, view the live demo

example

features

• editor
• native mcp server
• wasm demo
• quantize-aware kernels

media

setup

install

# burn_synth CLI
cargo install burn_synth

# bevy_synth app
cargo install bevy_synth

# burn_synth MCP stdio server
cargo install burn_synth_mcp

usage:

# burn_synth: run image -> GLB synthesis
burn_synth mesh \
  --quality balanced \
  --input docs/output_chair_bg_removed.png \
  --output /tmp/chair.glb

# bevy_synth: launch interactive app
bevy_synth --quality balanced

# burn_synth_mcp: start MCP stdio server
burn_synth_mcp

# burn_synth_mcp: run source scene -> object images -> TRELLIS assets -> BSN once
burn_synth_mcp \
  --backend wgpu \
  --trellis-quality low \
  scene-build \
  --source-scene-path /path/to/source_scene.jpg \
  --output-dir tmp/runs/scene_e2e \
  --candidate-count 2 \
  --candidate-retry-attempts 2 \
  --batch-size 0 \
  --trellis-pbr true

# To apply into a running bevy_synth viewer, launch the viewer with the same
# control path and add --apply to the scene-build command.
bevy_synth --mcp-scene-control-path tmp/runs/scene_e2e/scene_commands.json
burn_synth_mcp \
  --scene-control-path tmp/runs/scene_e2e/scene_commands.json \
  --scene-status-path tmp/runs/scene_e2e/scene_commands.status.json \
  scene-build \
  --source-scene-path /path/to/source_scene.jpg \
  --output-dir tmp/runs/scene_e2e \
  --apply

upstream

• see docs/SETUP.md for full setup and model import/runtime details
• supported models:
  • foreground: RMBG-1.4, RMBG-2.0
  • synthesis: TripoSG, Trellis2

note: pre-trained model weights have separate license

hardware recommendation

• 16GB VRAM

references

• assembler
• burn_dino
• lattice
• midi
• trellis
• triposg