Comfy-Org Drops Idol Into ComfyUI to Turn Photos Into 3D in Seconds

Speridlabs' open image-to-3D model Idol lands in ComfyUI, turning one picture into a million-Gaussian splat on 7.2 GB of VRAM.

·
·
·
Comfy-Org Drops Idol Into ComfyUI to Turn Photos Into 3D in SecondsPRO
  • Comfy-Org released a ComfyUI-ready repackage of Speridlabs' Idol image-to-3D model.
  • Idol turns one image into a 3D Gaussian splat with up to 1,048,576 Gaussians.
  • Runs on consumer GPUs in under 7.2 GB of VRAM, generates 1M splats in roughly 3 seconds on high-end hardware.
  • Built on a 600M parameter 3D Diffusion Transformer using DINOv3 and FLUX.2 VAE for image features.
  • Apache 2.0 license covers both weights and the idol-sdk code.
  • Slots into existing TripoSplat ComfyUI workflows with three file drops into diffusion_models, clip_vision and vae folders.

Comfy-Org has published four ready-to-load safetensors files for running Speridlabs’ Idol image-to-3D model through existing ComfyUI TripoSplat workflows. The Comfy-Org Idol repository removes the need to convert the upstream weights or build a new graph around them.

Speridlabs’ Idol research page describes an open model that converts one photograph, render, or generated image into a 3D Gaussian splat. Hidden surfaces are predictions based on the model’s learned shape priors, since a single view contains no direct information about the back or occluded sides. Outputs range from 32,768 to 1,048,576 Gaussians, and the code and weights use the Apache 2.0 license.

Inside the four-file drop

The repository packages a fine-tuned Idol variant for ComfyUI’s existing TripoSplat nodes. Copy each file into the corresponding directory under the ComfyUI installation:

  • Copy idol_fp16.safetensors to ComfyUI/models/diffusion_models/.
  • Copy dino_v3_vit_h_bf16.safetensors to ComfyUI/models/clip_vision/.
  • Copy flux2-vae.safetensors to ComfyUI/models/vae/.
  • Copy idol_decoder_fp16.safetensors to ComfyUI/models/vae/.

A recent ComfyUI installation with a working TripoSplat workflow should recognize the files after a restart. Existing graphs can then select the Idol weights through their model loaders. The release consists of model files designed for those nodes, so users without a TripoSplat graph must install or load the compatible workflow first.

How Idol invents the hidden sides

Pro article

This story is for Pro members

You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.

Trending
  • No trending articles

Comments

avatar

Next Reads