Sean Hutchinson | Orrich Lawson | Stable Diffusion
By now, you’ve read a lot about generative AI technologies like Midjourney and Stable Diffusion that convert text input into images in seconds. If you’re anything like me, I immediately wondered how you could use that technique to slap the Mona Lisa or Captain America in the face. who is
I have a long history of putting things in the face. It used to be a painstaking process of finding or taking a photo with the right angle and expression, then using Photoshop to graft my face onto the original. I thought the results were silly but worth it, and the process took a lot of time. You can now paste whatever your heart desires.
In this walkthrough, you’ll install Stable Diffusion locally on your computer, train Dreambooth on your face, and generate so many photos of yourself that your friends and family will have a silly photo deluge. I will explain how to stop it. The whole process takes about 2 hours from start to finish. Most of my time is spent babysitting my Google Colab notebook while training images.
Before we start, a few caveats.
System spec
For this tutorial, I am working on a Windows computer with an Nvidia 3080Ti with 12GB VRAM. To run Stable Diffusion you need his Nvidia graphics card with at least 4GB of video RAM.stable diffusion can It runs on Linux systems, Macs with M1 or M2 chips, and AMD GPUs, and can generate images using only the CPU. However, these methods require some tweaking, so a Windows machine with an Nvidia GPU is recommended for the purposes of this walkthrough.
ethical concerns
You should be aware that there are ethical concerns regarding image generation programs like Stable Diffusion. There are legitimate questions about how the data used to train Stable Diffusion was collected and whether it is ethical to train the program on an artist’s work without their consent. This is a large topic outside the scope of this walkthrough. Personally, I use Stable Diffusion as author for quick character sketches. This has become a very valuable part of my process. However, I don’t think there is a need to commercialize the work produced by Stable Diffusion, at least not until we have resolved the ethical dilemma and decided how to compensate potentially exploited artists. And for the time being, I think Stable Diffusion should be for personal use only.
Finally, technology like Stable Diffusion is both exciting and scary. It’s exciting because it gives people like me, who peaked artistically in finger painting in kindergarten, the ability to create the images I imagine. It’s scary because it can be used to create deepfakes that can Therefore, Stable Diffusion should only be trained on photos of yourself or consenting parties. period.
Well, who is ready to do this?
