auto_diffusers

CONTENTS

• About The Project
• How to use
• Description
• License
• Acknowledgement

About The Project

Enhance the functionality of diffusers.

• Search models from huggingface and Civitai.

How to use

pip install --quiet auto_diffusers

from auto_diffusers import EasyPipelineForText2Image

# Search for Huggingface
pipe = EasyPipelineForText2Image.from_huggingface("any").to("cuda")
img = pipe("cat").images[0]
img.save("cat.png")


# Search for Civitai
pipe = EasyPipelineForText2Image.from_civitai("any").to("cuda")
image = pipe("cat").images[0]
image.save("cat.png")

---

from auto_diffusers import (
    search_huggingface,
    search_civitai,
) 

# Search Lora
Lora = search_civitai(
    "Keyword_to_search_Lora",
    model_type="LORA",
    base_model = "SD 1.5",
    download=True,
    )
# Load Lora into the pipeline.
pipeline.load_lora_weights(Lora)


# Search TextualInversion
TextualInversion = search_civitai(
    "EasyNegative",
    model_type="TextualInversion",
    base_model = "SD 1.5",
    download=True
)
# Load TextualInversion into the pipeline.
pipeline.load_textual_inversion(TextualInversion, token="EasyNegative")

Description

Arguments of EasyPipeline.from_huggingface

Name	Type	Default	Input Available	Description
pretrained_model_or_path	str or os.PathLike	ー	ー	Keywords to search models
checkpoint_format	string	"single_file"	single_file, diffusers, all	The format of the model checkpoint.
pipeline_tag	string	None	ー	Tag to filter models by pipeline.
torch_dtype	str or torch.dtype	None	ー	Override the default torch.dtype and load the model with another dtype. If "auto" is passed, the dtype is automatically derived from the model's weights.
force_download	bool	False	ー	Whether or not to force the (re-)download of the model weights and configuration files, overriding the cached versions if they exist.
cache_dir	str, os.PathLike	None	ー	Path to a directory where a downloaded pretrained model configuration is cached if the standard cache is not used.
token	str or bool	None	ー	The token to use as HTTP bearer authorization for remote files.

model_format

Argument	Description
single_file	Only single file checkpoint are searched.
diffusers	Search only for `multifolder diffusers format checkpoint

Other_Arguments

Name	Type	Default	Input Available	Description
proxies	Dict[str]	None	ー	A dictionary of proxy servers to use by protocol or endpoint.
output_loading_info	bool	False	ー	Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
local_files_only	bool	False	ー	Whether to only load local model weights and configuration files or not.
revision	str	"main"	ー	The specific model version to use.
custom_revision	str	"main"	ー	The specific model version to use when loading a custom pipeline from the Hub or GitHub.
mirror	str	None	ー	Mirror source to resolve accessibility issues if you’re downloading a model in China.
device_map	str or Dict[str, Union[int, str, torch.device]]	None	ー	A map that specifies where each submodule should go.
max_memory	Dict	None	ー	A dictionary device identifier for the maximum memory.
offload_folder	str or os.PathLike	None	ー	The path to offload weights if device_map contains the value "disk".
offload_state_dict	bool	True	ー	If True, temporarily offloads the CPU state dict to the hard drive to avoid running out of CPU RAM.
low_cpu_mem_usage	bool	Depends on torch version	ー	Speed up model loading only loading the pretrained weights and not initializing the weights.
use_safetensors	bool	None	ー	If set to None, the safetensors weights are downloaded if they're available and if the safetensors library is installed.
gated	bool	False	ー	A boolean to filter models on the Hub that are gated or not.
kwargs	dict	None	ー	Can be used to overwrite load and saveable variables.
variant	str	None	ー	Load weights from a specified variant filename such as "fp16" or "ema".

---

Tip

If an error occurs, insert the token and run again.

Arguments of EasyPipeline.from_civitai

Name	Type	Default	Input Available	Description
search_word	string	ー	ー	Keywords to search models
model_type	string	Checkpoint	Details	The type of model to search for.
base_model	string	None	ー	Trained model tag (example: SD 1.5, SD 3.5, SDXL 1.0)
download	bool	False	ー	Whether to download the model.
force_download	bool	False	ー	Whether to force the download if the model already exists.
cache_dir	string, Path	None	ー	Path to the folder where cached files are stored.
resume	bool	False	ー	Whether to resume an incomplete download.
token	string	None	ー	API token for Civitai authentication.
skip_error	bool	False	ー	Whether to skip errors and return None.

search_word

Type	Description
keyword	Keywords to search model
url	Can be any URL other than huggingface or Civitai.
Local directory or file path	Search for files with the extensions: .safetensors, .ckpt, .bin
huggingface path	The following format: < creator > / < repo >

model_type

Input Available
Checkpoint, TextualInversion, Hypernetwork, AestheticGradient, LORA, Controlnet, Poses

License

In accordance with Apache-2.0 license

Acknowledgement

I have used open source resources and free tools in the creation of this project.

I would like to take this opportunity to thank the open source community and those who provided free tools.