Idol

What is Idol?

Idol is a model released by speridlabs that generates a 3D Gaussian Splat (3DGS) from a single image.

Idol has the same architecture as TripoSplat, so in ComfyUI it runs on the TripoSplat nodes.

Downloading the models

• background_removal
  • birefnet.safetensors (444 MB)
• diffusion_models
  • idol_fp16.safetensors (741 MB)
• clip_vision
  • dino_v3_vit_h_bf16.safetensors (1.68 GB)
• vae
  • flux2-vae.safetensors (336 MB)
  • idol_decoder_fp16.safetensors (576 MB)

📂ComfyUI/
└── 📂models/
    ├── 📂background_removal/
    │   └── birefnet.safetensors
    ├── 📂clip_vision/
    │   └── dino_v3_vit_h_bf16.safetensors
    ├── 📂diffusion_models/
    │   └── idol_fp16.safetensors
    └── 📂vae/
        ├── flux2-vae.safetensors
        └── idol_decoder_fp16.safetensors

image to 3DGS

• Remove Background
  • Removes the background with BiRefNet.
• TripoSplat Preprocess Image
  • Places the subject cut out by the mask in the center of a square with a black background.
• TripoSplat Decode
  • num_gaussians is the number of Gaussians to generate.
  • Going above 262144 only makes it denser; it does not add detail.
• Create 3D File (from Splat) → Save 3D Model
  • Saves the splat as .ply, and you can check it right away in the 3D view.