Summary

The web content provides a detailed guide for running the Stable Diffusion AI image generator on a local machine without requiring programming skills.

Abstract

The article titled "Run Stable Diffusion AI At Home — No Code Guide" explains how to set up and run the Stable Diffusion AI model, which generates images from text prompts, on a personal computer equipped with an NVIDIA GPU. It outlines the necessary hardware prerequisites, such as a minimum of 4GB VRAM and 10GB of disk space, and walks the reader through downloading and installing essential software like Anaconda, Git, and the Stable Diffusion model weights. The guide includes steps for setting up a Python environment, downloading the model repository, activating the environment, preloading necessary machine learning models, and finally, running the model to generate images. The author, Jim Clyde Monge, provides visual aids and personal experiences, including the time taken to generate an image and the hardware used. The article concludes with an invitation for readers to experiment with the model's parameters and share their results, as well as a note about an upcoming guide for running Stable Diffusion with a graphical user interface (GUI).

Opinions

The author, Jim Clyde Monge, believes that running Stable Diffusion locally is accessible to individuals without programming experience, as the guide is designed to be clear and comprehensive.
The author expresses gratitude and gives credit to the developer lstein for providing a modified version of the Stable Diffusion source code, indicating a collaborative and appreciative stance within the tech community.
Jim Clyde Monge encourages reader interaction and support by inviting comments, sharing of artwork, and suggesting readers sign up for a Medium membership to access more content from him and other writers.
The author's enthusiasm for the topic is evident through the inclusion of personal results and the anticipatory mention of a future guide, showcasing a commitment to ongoing learning and sharing within the field of AI-generated art.

Run Stable Diffusion AI At Home — No Code Guide

Recently, startup StabilityAI announced the release of Stable Diffusion, a powerful AI image generator that can now run on standard graphics cards.

Note: You don’t need any programming experience to follow along, it’s all spelled out.

Pre-requisites

Before doing the steps below, make sure you have all the requirements to run the AI model in your local hardware.

NVIDIA GPU with at least 4GB VRAM
At least 10GB of space in your local disk

Now on to the steps.

Step 1. Setup Git and Python environment

Download and install the latest Anaconda Distribution here.
Download and install the latest Git here.

Step 2. Download the model weights

Click on this link and download the latest Stable Diffusion library.

As of this writing, the latest version is v1.4-original.

Under the “Files and versions” tab, click on the checkpoint file and download it.

Note: The file is more than 4GB, so while you wait, go get a coffee or do something else.

Step 3. Download Stable Diffusion Repository

Download the source project from this GitHub repository.

The repository is called Stable Diffusion Dream Script, it’s a fork of the original source code modified by lstein. Shout out to that awesome guy.

Extract the files into your local. Inside the “ldm” folder, create another folder named “stable-diffusion-v1”. Like so in this screenshot.

Copy the model file sd-v1–4.ckpt we downloaded in Step#2 and paste it into the stable-diffusion-v1 folder.

Finally, rename the checkpoint file to model.ckpt.

Step 4. Activate the environment

Open up Anaconda cmd prompt and navigate to the “stable-diffusion-main” folder.

Now, we need to activate a few python packages.

(base) ~/stable-diffusion-main$ conda env create -f environment.yaml
(base) ~/stable-diffusion-main$ conda activate ldm
(ldm) ~/stable-diffusion-main$

The first line should download all necessary dependencies and packages to run the model. Note that files are a few gigabytes big so it might take a while to finish downloading everything.

Next, activate ldm by the conda activate ldmcommand.

You should see your command prompt prefixed by “ldm” like below.

Finally, load a couple of small ML models required by Stable Diffusion.

(ldm) ~/stable-diffusion-main$ python scripts/preload_models.py

Make sure you see that “success” message before we continue.

Step 5. Run the model

Now we can start generating images.

(ldm) ~/stable-diffusion-main$ python scripts/dream.py

The command line will pause on “dream>” to ask you to input your desired image prompt.

I am running on NVIDIA GeForce RTX 3060 Ti with 8GB RAM. The whole process took 12.82s to generate a 512x512 image.

Here’s the result.

It’s not too bad, right?

Okay, I guess that’s about it. In the next few days, I’ll do my best to write another article that shows how you can change the image by playing with the arguments.

You can change things like the resolution, the seed, the number of images made, and so on.

Try it out for yourself. Enjoy!

If you have any problems, aren’t sure how to do something, or want to show me your art, let me know in the comments.

[Update: 09/12/2022]

I published a new guide on how to run Stable Diffusion with GUI here. The steps are more straightforward and the repo is an optimized version of the original.

If you like to read more stuff like this in Medium, consider supporting me and thousands of other writers by signing up for a membership.

Join Medium with my referral link - Jim Clyde Monge

Read every story from Jim Clyde Monge (and thousands of other writers on Medium). Your membership fee directly supports…

jimclydemonge.medium.com