Skip to content

Model Storage

Root Path

Default root path for all models in sdnext is models folder inside SD.Next installation folder
To change the root path, set it in: Settings -> System Paths -> Root model folder
All individual folder locations can be customized in: Settings -> System Paths

Tip

For details on Model loading, including how to use model fine-tunes,
see Model Loading docs

Model Folders

  • Folder with stable diffusion models
    Use: SD/SDXL single-file models
    Default: <models>/Stable-diffusion
  • Folder with Huggingface models
    Use: Models downloaded from Huggingface
    Default: <models>/Diffusers
  • Folder for Huggingface cache
    Use: Shared model components for models downloaded from Huggingface, for example large text-encoder can be shared between multiple models
    Default: <models>/huggingface
  • Folder for XET cache Use: BLOBs for model components
    Default: <models>/xet

Note

If you want to share your SD/SDXL models between different applications,
configure Folder with stable diffusion models accordingly

Important

Manually deleting a Diffusers/Huggingface model or model component
may not remove all associated files due to caching and shared components
To reclaim storage space after deleting, use:
right panel -> Models -> Prune Cache

Warning

Manually downloading models and placing them in Diffusers or huggingface folders is not supported
Those folders are automatically populated by SD.Next when models are downloaded through the application

Model Fine-tunes

  • Folder with UNET files
    Use: Unet/DiT/Transformer fine-tunes that can be applied to the base model
    Default: <models>/UNET
  • Folder with Text encoder files
    Use: Text encoder fine-tunes that can be applied to the base model
    Default: <models>/Text-encoder
  • Folder with VAE files
    Use: VAE fine-tunes that can be applied to the base model
    Default: <models>/VAE

Other Models

  • Folder with LoRA network(s)
    Use: LoRA networks that can be specified in the prompt and loaded during processing
    Default: <models>/Lora
  • Folder with textual inversion embeddings
    Use: Textual inversion embeddings that can be specified in the prompt and loaded during processing
    Default: <models>/embeddings
  • Folder with Yolo models
    Use: Yolo models are used for object detection tasks
    Default: <models>/Yolo
  • Folder with Upscaler chaiNNer models
    Use: chaiNNer models are used for image upscaling tasks
    Default: <models>/chaiNNer
  • Folder with Upscaler Spandrel models
    Use: Spandrel models are used for image upscaling tasks
    Default: <models>/Spandrel

Control Models

  • Folder with all Control models
    Default: <models>/control
  • <control>/processor: image pre-processing models
  • <control>/controlnet: ControlNet models
  • ...

Other Folders

  • Folder with user-defined styles
    Default: <models>/styles
  • Folder with user-defined wildcards
    Default: <models>/wildcards
  • Folder with tag autocomplete files
    Default: <models>/tags