TongFlow Now Integrates TripoSplat — One Image to a 3D Gaussian Splat

TongFlow now integrates TripoSplat. The official tongflow-modal-triposplat plugin turns a single image into a 3D Gaussian splat — drop a photo or render on the canvas and get back a .splat you can orbit, zoom, and export, GPU-powered via Modal.

What you get

TripoSplat (VAST-AI-Research/TripoSplat, by TripoAI) is a feed-forward model that reconstructs a 3D Gaussian splat from one 2D image. The plugin wires it up as a single ready-to-run node:

• Image → 3D Model (image-gen-model) — feed in one image, get back a 3D Gaussian splat.

Background removal is built in, so any photo, illustration, or generated frame works — no manual masking. You control the Gaussian count (up to 262,144), trading visual quality against rendering and storage cost. The output is a standard .splat, ready for any Gaussian-splat viewer.

See it on the canvas — new in v0.1.2

You no longer have to leave TongFlow to inspect a result. As of v0.1.2, the 3D model node renders Gaussian splats in-app via SparkJS: orbit with a free-tumble trackball camera (no pole lock), zoom, and pan — right on the canvas, both in the node preview and the fullscreen viewer. Rendering is on-demand, so even a 262K-splat asset stays smooth. Want it elsewhere? Export the .splat / .ply and open it in SuperSplat or Spark.

Built to fit one GPU

The plugin vendors TripoSplat’s compact, near-zero-dependency pipeline and runs it on a single A100-40GB via Modal. Weights are pulled once to a shared volume; after that a single image takes a couple of minutes end to end, cold-start model load included. Tuning knobs — Gaussian count, sampler steps, guidance — live in the plugin README.

Start for free

The plugin runs on Modal, which gives every new account up to $30/month of free GPU compute on real A100-class hardware — enough to splat plenty of images without owning a GPU. And because the VAST-AI/TripoSplat weights are public (MIT), you only need two credentials in TongFlow Settings: MODAL_TOKEN_ID and MODAL_TOKEN_SECRET. No gated-model access to request.

Try it

1. Install the plugin — open the plugin manager in TongFlow and install tongflow-modal-triposplat. Usable immediately, no restart.
2. Add your Modal tokens in Settings.
3. Drop an image on the canvas, connect an Image → 3D Model node, and run. The platform auto-deploys to Modal on first use, and the result renders right there.

New to TongFlow? Start with the v0.1.0 release post, grab the desktop app from Releases, or jump into the hosted studio at app.tongflow.com. If this is useful, a star on GitHub helps a lot.

Expand your imagination, stretch your ideas — give it a try.