mirror of
https://github.com/vladmandic/automatic
synced 2026-08-25 22:20:46 +02:00
Page:
Loader
Pages
AMD MIOpen
AMD ROCm
API
Advanced Install
Benchmark
CHANGELOG
CLI Arguments
CLI Tools
CLiP Skip
Caption
Control HowTo
Control Settings
Control Technical
Debug
Detailer
Dev AICoding
Dev Docs
Dev Extensions
Dev GettingStarted
Dev Hints
Dev Home
Dev Locale
Dev SelfStart
Dev Structure
Dev Theme
Dev UI
DirectML
Docker
Enso
FAQ
FLUX
Features
FramePack
Gated
Getting Started
Google GenAI
HiDream
Home
Hotkeys
HuggingFace
IPAdapter
Ideogram
Installation
Intel ARC
Kanvas
LTX
Launcher
LoRA Technical
LoRA
Loader
MacOS Python
Malloc
MiniMax
Model Loading
Model Support
Models Tab
Models
Networks Search
Networks
Notes
NudeNet
Nunchaku
ONNX Runtime
Offload
OpenVINO
Outpaint
Parameters
Performance Timers
Performance Tuning
Platforms
Process
Profiling
Prompt Enhance
Prompting
Python
Quantization
Reprocess
SD Pipeline How it Works
SD Training Methods
SD XL
SD3
SDNQ Quantization
Schedulers
Scripts
Stability Matrix
Stable Cascade
Styles
Themes
Troubleshooting
Update
Using LCM
VAE
Video
WSL
Wildcards
XYZ Grid
ZLUDA
_ToDo
index
nVidia
Clone
4
Loader
Vladimir Mandic edited this page 2026-05-13 12:33:44 +02:00
Table of Contents
Custom Model Loader
The custom model loader is available in Models -> Loader. You can use it to load any supported diffusion model with default or custom components.
Model Type
- Can be any currently supported model type.
- Autodetect attempts to load a repo for any model. This is useful for new Hugging Face models that are not yet in the predefined list.
- Current replaces selected components without reloading the model itself.
Model Repo
A valid model repo on Hugging Face is required to access the base model config.
- Default model repo is provided for common models
- Model repo can be overridden to any valid repo on Hugging Face
Loadable Components
Note
Any model component marked as loadable can be overridden. If neither local nor remote values are set, the default from the model repo is used. Any loadable component with a set value is loaded from that value.
Local vs Remote
- You can use local or remote models.
- A local model can be a valid path to a safetensors file or a folder with HF-style structure.
- A remote model can be any valid Hugging Face repo, with or without a subfolder, or a full URL to a Hugging Face-hosted file.
Dtype
- Overrides Torch compute dtype when loading each component.
Quant
- Enables quantization when loading each component.
- Quant settings are taken from Settings -> Quantization.
Recipe
Model recipes can be saved to a JSON file for later use.
Warning
Load/save recipe is not yet implemented.