Skip to content
eduardoabreu81 edited this page Nov 6, 2026 · 1 revision

Models

The files you need

Part File Size Folder
Checkpoint idol_fp16 741 MB models/Stable-diffusion
Image encoder (DINOv3 ViT-H+/16) dino_v3_vit_h_bf16 1.7 GB models/text_encoder
FLUX.2 VAE flux2-vae 336 MB models/VAE
Idol decoder idol_decoder_fp16 576 MB models/VAE

These come from the Comfy-Org repack, which needs no Hugging Face account. The original image encoders are in gated repositories; the repack avoids that.

The extension recognizes each file from the tensor names in its header, so the file names do not matter.

Downloaded by themselves

Into models/Idol, the first time they are needed. To install them by hand, put them there with these names.

File Size When
birefnet.safetensors 444 MB A picture without transparency
idol_turbo_slow_comfy.safetensors 59 MB Idol Turbo
idol_turbo_fast_comfy.safetensors 59 MB Idol Turbo

Other files that are recognized

  • The release files of speridlabs/idol: denoiser as the checkpoint, and decoder with decoder_gs as the decoder. Recognized by their tensor names; the runs on this page's results used the Comfy-Org files.
  • The stock TripoSplat denoiser has the same layout and is recognized too, but it has not been run.

Not supported yet

  • GGUF files of Idol, such as TimeLordRaps/Idol-GGUF
  • Forge's Diffusion in Low Bits options on the Idol checkpoint have not been tested

Licenses

Idol and Idol Turbo are under the Apache License 2.0. DINOv3 and the FLUX.2 VAE have licenses of their own: see their pages before commercial use.

Clone this wiki locally