VisForge

VisForge is a playground for building a ComfyUI-inspired visual node editor with an Unreal Engine aesthetic. The current milestone focuses on a rich React + Rete.js frontend, Babylon.js powered previews, and a FastAPI backend skeleton that can accept serialized graphs for future execution.

Frontend (Rete.js + Babylon.js)

Features

• Unreal‑style node styling with collapsible bodies and floating connection orbs.
• Node library sidebar grouped by category for quick drag-free placement.
• Workflow tabs with local persistence (save, save‑as, rename) backed by localStorage.
• Prebuilt nodes: Load Image, Load Model, Separate Channels, Combine Channels, Show Image, Preview 3D, Generate Tripo Model, Generate Hy 2.1 Model, Remove Background, Save Model, and Save Image.
• New Generate Tripo Model node that submits an image to the backend TripoSG pipeline and returns a GLB for downstream preview.
• Added Generate Hy 2.1 Model node backed by the Hunyuan3D-2.1 image-to-3D pipeline with configurable MC/DMC surface extraction.
• Background cleanup can be done inline via the Remove Background node before handing images to generation models.
• New Remove Background node wraps rembg with RGB/RGBA/transparent or custom-color compositing for fast cleanup workflows.
• Image pipeline helpers that split/merge RGBA channels directly in the browser.
• Babylon.js viewport with switchable Base, Wire, and Norm shading modes and smarter GLB/GLTF loading.

Getting Started

cd frontend
npm install
npm run dev

The dev server runs on http://localhost:5173. The first load may take a few seconds because Babylon.js is bundled.

Optionally create a .env.local in frontend/ with VITE_BACKEND_URL=http://localhost:8000 so the TripoSG node points at the FastAPI server during development.

Backend (FastAPI)

A lightweight execution service is scaffolded at backend/app/main.py. It currently validates graphs and returns a topological execution plan, streaming placeholder updates over a WebSocket.

Setup

cd backend
python -m venv .venv
source .venv/bin/activate  # On Windows use: .venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Note: The TripoSG node relies on heavy ML dependencies (PyTorch, diffusers, pymeshlab, etc.) and the initial run downloads the required checkpoints into backend/repos/TripoSG/weights/. Ensure you have sufficient disk space and (optionally) a CUDA-capable GPU for reasonable performance. Note: The Hunyuan3D-2.1 node pulls the shape-generation weights from Hugging Face into backend/repos/Hunyuan3D-2.1/weights/ on first use. The download is large; plan for ample disk space and a GPU if you want practical inference times. The OpenAPI docs will be available at http://localhost:8000/docs once the server is running.

Project Layout

VisForge/
├─ frontend/          # Vite + React + Rete.js + Babylon.js client
├─ backend/           # FastAPI service skeleton
└─ README.md          # This file

Next Steps

1. Implement secure node execution inside the FastAPI service (e.g., subprocess sandbox or worker queue) and bridge it with the frontend WebSocket events.
2. Persist graphs and assets on the backend (e.g., PostgreSQL or SQLite + blob storage).
3. Add authentication/authorization before exposing execution endpoints publicly.
4. Extend node catalog (texture transforms, lighting, procedural nodes) and support undo/redo via Rete plugins.

Happy forging!