A workflow studio that treats AI as a modality transform

What it does

TongFlow is a visual workflow studio for building multi-modal generative AI pipelines. It wraps AI capabilities—text generation, image synthesis, video editing, audio separation, even 3D extraction—into nodes that you arrange on a canvas. The desktop app is a thin (~10 MB) shell around a cloud studio; a fully local, account-free version requires running your own instance.

The interesting bit

The conceptual reduction is the clever part: every model is treated as a modality transform, and the entire interface collapses into just three operations—add, transform, and combine. That is a genuine attempt to abstract away the usual maze of model-specific parameters and Python glue. It also ships with a broad plugin ecosystem covering everything from lip-sync and motion capture to open-vocabulary sound separation and music arrangement.

Key highlights

• Supports an unusually wide modality stack: text, image, audio, video, documents, URLs, and 3D models.
• Plugin architecture splits the work between API gateways (OpenAI, Gemini, Kling, etc.) and GPU/CPU workers running on Modal.
• The “three operations” UX (add, transform, combine) aims to replace manual node-wiring and parameter tuning.
• Includes niche AV utilities often missing from no-code tools: shot detection, subtitle removal, watermark removal, stem separation, and motion-capture-to-GLB.
• AGPL-3.0 licensed; official plugins are open and extensible.

Caveats

• The desktop app requires signing in via Google or WeChat and relies on the cloud studio; going fully local means running your own instance.
• macOS builds are currently not notarized, so Gatekeeper will flag the app as damaged on first launch.
• Several listed nodes (speaker diarization, voice/timbre replacement) exist on the canvas but lack official plugins yet.

Verdict

Worth a look if you build media pipelines that hop between text, image, audio, and video and you are tired of maintaining one-off scripts. Skip it if you want a fully offline, batteries-included desktop experience out of the box—the cloud dependency and plugin setup are real.