Hacker News
Graviton is a new tool that transforms ComfyUI into a production-ready workflow engine for generative AI. It allows users to chain complex workflows, add approval gates, handle retries, and distribute processing across multiple GPUs, with automatic API export capabilities.
AI 生成摘要
Graviton 是一款新工具，可將 ComfyUI 轉變為生產級的生成式 AI 工作流程引擎。它允許使用者串聯複雜的工作流程、加入審核閘、處理重試，並將處理分散到多個 GPU，同時具備自動匯出為 API 的功能。
We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Graviton: Daisy-Chain ComfyUI workflows. Distribute among multiple GPUs
Orchestrate multi-step ComfyUI workflows with approvals, retries, and distributed GPU load balancing.
Graviton turns ComfyUI into a production-ready workflow engine. Chain multiple workflows together, add human approval gates, automatically retry failed steps, and distribute work across multiple GPU servers.
Copy graviton/folder_paths.py to your ComfyUI installation directory:
This enables Graviton to sync workflows from your ComfyUI server.
Export your ComfyUI workflows (API format) and place them in your ComfyUI server's models/templates/ directory:
Graviton will automatically sync these workflows from your ComfyUI server.
Edit config.yaml with your ComfyUI server address (see config file for examples).
That's it! Open http://localhost:3000 to access the visual editor.
For best compatibility, your ComfyUI workflows should follow these guidelines:
When Graviton syncs a workflow, it auto-generates an _overrides.json file in the templates/ directory:
To customize which parameters are editable in the UI:
To check if a workflow will work on a specific ComfyUI server:
This checks if the server has all required custom nodes installed.
Use Cloudflare Tunnel or ngrok to expose your remote ComfyUI:
See config.yaml.example for more options.
When a step has approval enabled:
This enables iterative refinement without restarting the entire chain.
MIT
Graviton: Daisy-Chain ComfyUI workflows. Distribute among multiple GPUs
There was an error while loading. Please reload this page.
0
2
