The creator broke the model into 20 CoreML models so as not to run out of memory.
With Stable Diffusion quickly becoming the most widely used text-to-image AI on the market, most likely thanks to its code being available for free, it is no wonder that more and more tinkers start experimenting with it, sharing mind-blowing results in the process. iOS Developer and Programmer Matt Waller, for example, recently presented an absolutely astonishing feat that impressed the entire Internet, the coder managed to run SD directly on a 4-year-old iPhone XS.
According to Matt, the phone version of the AI requires around 5 minutes to generate one iteration of the prompt. To demonstrate its capabilities, the developer used the "pineapple on a white table" prompt and shared 20 CoreML models generated by the AI, which with each iteration resembled the description more and more closely. In the end, it took around an hour for Stable Diffusion to finish the job and produce a great-looking picture.
"Alright friends, I'm going to dive back into coding, but also, as I've been so preoccupied with whether I could, I'll stop to think if I should. What safeguards to add, that sort of thing," commented the developer.