Big Sleep AI: What It Is, And How To Use It

Have you ever seen a picture that doesn't make sense? One where you are unable to comprehend the items in the picture? Or maybe, an image like this one that just seems eerie and out of place? Chances are, those images were generated by an AI. Big Sleep is an AI made by Ryan Murdock, using open source resources from OpenAI. It is good at generating Van Gogh-style artwork, and sometimes real looking animals, but do not expect too much of it. Here are some images I generated myself using Big Sleep, one for nightmare, and one for fear. If you are curious as to how to make these on your own, I am here to tell you about how to use it, since there were no good guides out there and I myself was confused as fuck at first. (Sidenote: it is cool as fuck for making album covers.)

For the sake of this beginners tutorial (and the fact that the required python version won't run on my Debian Stable install), I will be showing you how to run Big Sleep from the Colab page. Note that this unfortunately requires you to sign in to a Google account, which I wind up doing after the first step. To begin, open the Colab page. It will look intimidating seeing all the lines of various code at first, which is what confused me, but it is actually really simple. Find and click the play button under "Restart the Kernel". From here, the script will start loading all the necessary dependencies. Wait for this to finish, you will know it is done after the play button re-appears. Once this is done, we need to tell Big Sleep what kind of image to generate. This can be done by pressing CRTL+F and typing "Van Gogh", as this is the default prompt for Big Sleep. I will be using the word "fear" for my image generation. After this, click "Runtime" on the menu bar, then "Restart and run all". Now, scroll down to the part of the page that says "Diagnostics", and your image will start generating right above it.

Easy as pie, right? I know I definitely was surprised at how needlessly complicated some tutorials I saw were making it, I nearly gave up running the darn thing myself because of that. Do note that the image generation is a lengthy process, and takes up to 200 frames/3 hours or so to generate the final image. However, this is not a requirement, and if you like anything that it generates before reaching frame 200, you can simply save the image and stop the process. I have used this tool to generate a few album covers for my music project, as well as saved some abstract artwork that it decided to make in between. I usually let the AI generate images for about 30 minutes, but I have gone as quickly as 5 and as long as 2 hours. Personally, I am usually against anything involving Artificial Intelligence, however, little things like this seem to be relatively harmless. well, aside from the insane amounts of nightmare fuel it can generate. Indeed, even with some of the more tame words I have tried, the images always seem to turn out scary, or at the very least, unsettling. Perhaps this is something that can change over time, using machine learning algorithms... or perhaps this is our look inside the "mind" which is artificial. Jokes aside, Big Sleep is pretty cool to me.

Some More Information

Ryan Murdock, also known as lucidrains, is the one who created this nifty tool. It uses both OpenAI, and BigGAN. This is also a good opportunity to state that I take no credit for creating this software whatsoever, I only thought that it was cool as shit, so I decided to share it here for others to read about. If you are interested, I have included a list of resources for more information about/ scary occurrences with/ lots of things involving artificial intelligence below.