Offline Execution of StableDiffusionPipeline from_single_file in Python

Question

I’m working with a Python script that uses the StableDiffusionPipeline from the diffusers module. Here’s the relevant part of my code:

from diffusers import StableDiffusionPipeline
MODEL_PATH = '/home/mypath/Stable-diffusion/limitlessvision_v20.safetensors'
pipeline = StableDiffusionPipeline.from_single_file(MODEL_PATH)

The line pipeline = StableDiffusionPipeline.from_single_file(MODEL_PATH) only works when I have an active internet connection. If I’m offline, I encounter the following error:

ConnectionError: HTTPSConnectionPool(host='raw.githubusercontent.com', port=443): Max retries exceeded with url: /CompVis/stable-diffusion/main/configs/stable-diffusion/v1-inference.yaml (Caused by NewConnectionError('<urllib3.connection.HTTPSConnection object at 0x7f27f1dc4b50>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution'))

Given that my .safetensors model file is already stored locally on my computer, I would like my code to function even when I’m offline. How can I achieve this? Any help would be appreciated. Thanks!

Answer 1

I ran into the same thing and stepped through the code a bit, which is actually less daunting that it seemed initially. Here is what I found:

Direct Answer to Question

You are missing the config file which is specified with the original_config_file parameter. That is why it is trying to download the file.

Some of the basic stable diffusion ones can be found here: https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion

But also the error should tell you the full path it's going to, so you can just download exactly the needed file.

I also recommend setting local_files_only=True just for good measure.

Additional Data

Once you have this running you mayfind you are missing two more items (this depends on your model, and the errors will give you details.)

You need to create a directory called openai/clip-vit-large-patch14 relative to your execution directory and download into it the contents of this: https://huggingface.co/openai/clip-vit-large-patch14/tree/main

Then by default it tries to download the safety checker model. You can similarly download it into the CompVis/stable-diffusion-checker-safety-checker directory, relative to your execution directory. You can get the files from here: https://huggingface.co/CompVis/stable-diffusion-safety-checker/tree/main

Or you can just set load_safety_checker=False, although this is not recommended.

Sample loading code

pipe = StableDiffusionPipeline.from_single_file(
   "./path/to/model",
   local_files_only=True
   original_config_file="./path/to/v1-inference.yaml"
   # load_safety_checker=False
)

Answer 2

You can execute the StableDiffusionPipeline from the torchdyn library to perform stable diffusion on a single file by following these steps:

1. First, install the necessary libraries:

pip install torchdyn
pip install torchaudio  # (if audio processing is required)

2. Next, import the required modules and classes:

import torch
from torchdyn.models import StableDiffusionPipeline
from torchvision import transforms

3. Load your single file data into a PyTorch tensor and apply any necessary preprocessing using transforms from torchvision. For example:

file_path = 'path_to_your_single_file'
data = your_custom_data_loading_function(file_path)

# Apply preprocessing if needed
preprocess = transforms.Compose([transforms.ToTensor(), your_additional_transforms])
processed_data = preprocess(data)

4. Once the data is preprocessed, create an instance of StableDiffusionPipeline and execute it with the processed data:

model = StableDiffusionPipeline()
output = model(processed_data)

Could you adapt this code to fit your specific use case and data format?

Feel free to ask if you have any issues or need further assistance!