TripoSplat Brings Native 3D Gaussian Splatting to ComfyUI

ComfyUI, the node-based interface for Stable Diffusion and generative media, integrated native support for 3D Gaussian Splatting through the TripoSplat model. This update allows creators to generate high-fidelity 3D representations from a single 2D image directly within the ComfyUI workspace. By moving away from external scripts and third-party wrappers, the platform now treats 3D data as a first-class citizen alongside images and video.

What's new

ComfyUI now includes a dedicated set of nodes for the TripoSplat model, a feed-forward transformer architecture designed for rapid 3D reconstruction. As of February 2025, the implementation allows users to input a single image and output a 3D Gaussian Splat (PLY file) in seconds. Unlike traditional photogrammetry which requires dozens of photos, TripoSplat uses a pre-trained model to predict spatial geometry and texture from a single perspective.

The update introduces a specialized 3D viewer node within the ComfyUI interface. This allows users to rotate, zoom, and inspect the generated 3D asset in real-time before exporting. The model is optimized for speed, generating a reconstruction in roughly 0.5 seconds on a standard consumer GPU, making it significantly faster than older NeRF (Neural Radiance Fields) approaches.

How it fits your workflow

ComfyUI users can now incorporate 3D asset generation into larger creative pipelines, such as generating a character image with Stable Diffusion and immediately converting it into a 3D splat for pre-visualization. This workflow serves as a local, open-source alternative to cloud-based services like Luma AI’s Genie or CSM.ai. For filmmakers and game developers, this provides a way to rapidly prototype background assets or digital doubles without manual modeling.

In a production environment, the TripoSplat nodes in ComfyUI compete with tools like Rodin or Meshy by offering more granular control over the generation process. Because it is node-based, an editor can chain the output of a 3D splat into a video generation node, using the 3D data to guide camera movement or depth maps. This level of integration is currently a distinct advantage over standalone 3D generators that require manual file transfers between different software packages.

What it costs / how to try it

ComfyUI users can access TripoSplat by updating their local installation and downloading the model weights from the ComfyOrg Hugging Face repository. The tool is free and open-source, though it requires a modern NVIDIA GPU with at least 8GB of VRAM for optimal performance.