OpenCartridge 将 ComfyUI 工作流封装为本地 Docker 容器,提供统一界面生成游戏资产,降低独立开发者使用本地 AI 的门槛。
🎮 Someone just built a completely local AI game asset studio that generates sprites, animations, sound, and 3D models, all from one interface. No subscriptions or cloud credits.
Meet OpenCartridge, a new open-source project by developer camenduru, built on TostUI and powered by local ComfyUI workers.
Instead of installing and configuring separate AI workflows for every asset type, OpenCartridge packages them into one-click Docker cartridges.
And everything runs on your own GPU.
🎨 What's inside?
▪️ Visual canvas: Layers, transforms, undo/redo, and asset composition
▪️ Sprite generation: Create 3×3 character sheets with eight isometric directions from a single reference
▪️ Sprite animation: Turn those eight views into synchronized walking animations
▪️ Pixel art: Transform characters into dark-fantasy pixel art inspired by Loop Hero
▪️ Background removal: Extract clean alpha mattes from images and videos using CorridorKey 2.0
▪️ Image-to-3D: Use Speridlabs' Idol to reconstruct objects from a single picture, including hidden surfaces
▪️ 3D viewer: Preview GLB assets and orbit around them
▪️ Audio tools: Waveform visualization and generation through compatible cartridges
▪️ Exports: PNG, MP4, WebM, and GLB
🧩 The cartridge system is the clever part
Each AI workflow lives inside its own Docker container.
Want to generate sprites? Install the sprite cartridge.
Need a 3D prop? Start the Idol cartridge.
Working with video? Activate CorridorKey.
Everything is managed directly from OpenCartridge:
▪️ Download and install workers with one click
▪️ Start, stop, or delete containers
▪️ See installation progress and worker status
▪️ Automatically connect workers to local storage
▪️ Load only the models needed for your current task
No manual Docker commands or ComfyUI node configuration for each cartridge.
The interface even includes a self-update mechanism that pulls the latest source, rebuilds the application, and restarts it.
🧠 A practical game development workflow
Imagine creating assets for an indie RPG.
1. Upload a reference image of your character.
2. Generate a 3×3 sprite sheet with eight isometric facings.
3. Animate the sheet into a looping walk cycle.
4. Create matching pixel-art variations.
5. Generate a 3D prop from a single image using Idol.
6. Preview the GLB model and export everything into your game engine.
All from the same browser interface, with generation happening locally.
⚙️ Requirements and installation
▪️ NVIDIA GPU: Approximately 24GB VRAM for the current workers
▪️ Docker Desktop: Required to run the application and its cartridges
▪️ Built-in MinIO: Persistent storage for uploaded files and generated assets
▪️ Browser interface: Available at localhost:3000
▪️ License: MIT
▪️ No account, API key, or generation credits
Initial installation still requires downloading Docker images and model weights, which can occupy several gigabytes.
The documentation currently provides a Windows AMD64 installation path. Other configurations may require additional setup.
⚠️ One important detail
OpenCartridge itself is free and MIT-licensed, but individual AI models and worker dependencies can have their own licensing terms.
Also, not every asset is automatically game-ready. Generated sprites may need cleanup, while 3D assets may require further optimization or rigging.
💡 Why this matters
Local generative AI has become incredibly powerful, but using it often means managing dozens of models, custom nodes, and disconnected interfaces.
OpenCartridge approaches the problem differently.
Instead of building another AI model, it builds a unified production environment around the models that already exist.
For indie developers and solo creators, that could make local AI asset generation considerably more practical.
And the cartridge architecture means new tools can be added without redesigning the entire studio.
One GPU. One interface. Multiple AI-powered asset pipelines.
🔗 GitHub: https://github.com/camenduru/OpenCartridge
#GameDev #ComfyUI #OpenSourceAI #GenerativeAI
来源:CuteGuy☣️ · x.com