Running Waifu Diffusion with energy saving vram on Windows


Setting up an environment for AUTOMATIC1111's repository in Windows environment

▼ Execution environment
Edition Windows 11 Pro
Version 21H2
OS build 22000.1098
Experience Windows Feature Experience Pack 1000.22000.1098.0

GPU: GTX 1060

1 Clone the repository

As in the previous article, the prerequisite is to create an appropriate git repository on Ubuntu and clone the repository.

> git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui

Also, download the model files from hugging face and put them under the models/stable-diffusion folder.
I used Waifu Diffusion v1.3 models with float16.

I believe you can download it from here

│ Put Stable Diffusion checkpoints here.txt
│ wd-v1-3-float16.ckpt

2 In my case, I use Anaconda to create a virtual environment to maintain my Python environment, so I install Anaconda Navigator for Windows.


3 Run Anaconda PowerShell Prompt

Anaconda PowerShell Prompt is installed with Anaconda Navigator by default.

If you feel that you are missing a package when running WebUI, try installing the package in the Anaconda environment and see if it works.

> pip install -r requirements.txt

4 Run webui.bat

> cd stable-diffusion-webui
> . /webui.bat
venv "C:\Users\keigo\git\stable-diffusion-webui\venvScripts\Python.exe"
Python 3.9.12 (main, Apr 4 2022, 05:22:27) [MSC v.1916 64 bit (AMD64)].
Commit hash: 1fbfc052eb529d8cf8ce5baf578bcf93d0280c29
Installing requirements for Web UI
Launching Web UI with arguments:
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Making attention of type 'vanilla' with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Working with z of shape (1, 4, 32, 32) = 4096 dimensions. making attention of type 'vanilla' with 512 in_channels
Loading weights [84692140] from C:\Users\keigo\git\stable-diffusion-webui\models\Stable-diffusion\wd-v1-3-float16.ckpt
Global Step: 683410
Applying cross attention optimization (Doggettx).
Model loaded.
Loaded a total of 0 textual inversion embeddings.
Running on local URL:

To create a public link, set `share=True` in `launch()`.

5 The service runs on, so access it, put in the prompt and draw!

It took about 20-30 seconds to generate the image.

6 vram during drawing consumed about 4.3GB!

It seems that a model with float16 precision can be used to generate a 512x512 image on a GTX 1060 GPU with about 6GB of vram.



Thanks to everyone behind these awesome projects,
First I would like to contribute the project by introducing the best environment run with the user PC environment.
And I will contribute the democratized AI through this activity.


Translated with www.DeepL.com/Translator (free version)