tong-io/tongflow

TongFlow – Open‑Source Multi‑Modal GenAI Workflow Studio

What it is

• A visual “studio” where you drag‑and‑drop nodes that wrap generative AI models. Each node performs a simple operation – add (bring in data), transform (run a model), or combine (merge results). The system treats every model as a modality transform (e.g., text→image, audio→text) and lets you chain them together without writing code.

Key capabilities

Category	Example nodes (official plugins)
Add	Text, image, sketch, audio, video, document, URL, 3‑D model, camera/photo capture
Transform	• Text generation / rewriting
• Image generation, editing, upscaling, pose detection, matting, segmentation, normals
• Video generation, image‑to‑video, interpolation, editing, upscaling, subtitle/watermark removal
• Audio: music generation, speech synthesis, transcription, noise‑reduction, source‑separation, stem extraction
Combine	Image fusion, lip‑sync, emotive speech, character swap, motion transfer, text merging
Helpers	Clip concatenation, mux/demux, shot detection, text chunking, batch grouping
Other	Image → 3‑D, video → motion‑capture, document → text, web‑page → text

How you run it

1. Desktop app – a ~10 MB installer (macOS DMG or Windows MSI) that launches a thin client pointing at the cloud studio at app.tongflow.com. Sign‑in with Google or WeChat.
2. Fully local – you can self‑host the backend (source or Docker) and run the plugins on your own compute. The repo provides a Run from source and Run with Docker guide, plus a “Self‑host setup” for plugging in your own API keys or Modal tokens.

Plugin ecosystem

• API plugins expose commercial model providers (Google Gemini, OpenAI, DeepSeek, ByteDance, xAI, Runway) with a per‑node model picker.
• Router plugins aggregate many third‑party models behind a single node (OpenRouter, CometAPI, ToAPIs, APIMart, Replicate, fal.ai).
• GPU/CPU plugins run on Modal (or locally) and implement the heavy‑weight media processing: image generation (Z‑Image, ERNIE, Krea2, Flux‑2), video generation (MiniMax‑H3, Bernini‑R), audio/music suites (ACE‑Step, LeVo, MiniMax‑Music3), vision tools (SAM‑3, Sapiens2, SenseNova), OCR/document extraction, and more.

Why it matters

• Low barrier, high ceiling – you only need to place three kinds of nodes; the UI hides model‑specific parameters.
• Multi‑modal – supports text, images, video, audio, 3‑D, and documents in a single canvas.
• Open and extensible – anyone can publish a new plugin that follows the same node interface; the core stays lightweight while the ecosystem can grow indefinitely.

Installation & licensing

• Install the desktop client from the GitHub Releases page (macOS universal dmg, Windows MSI). macOS builds currently require clearing the quarantine flag (xattr -cr).
• The project is released under AGPL‑3.0, so any modifications that are deployed as a service must also be open‑sourced.

Where to get help

• Discord community: https://discord.gg/K7V8az94Zf
• Full documentation and multilingual READMEs are in the docs/ folder.

---

TL;DR: TongFlow is an open‑source, visual workflow platform for chaining together generative AI models across many modalities. It ships a lightweight desktop front‑end, a cloud‑hosted studio, and a plug‑in system that connects to dozens of commercial and open‑source model APIs, letting users build anything from a simple text‑to‑image prompt to a full music‑video production without writing code.