@Magnus919 7/ I think that’s it. Many things don’t work yet on the M1 architecture (like upscalers) so I have limited possibilities to experiment for now. I’d also like to sleep, occasionally.

I hope it helps.

@Magnus919 2/ I usually generate on a MBP M1 Pro with AUTOMATIC1111 WebUI. I also like the InvokeAI fork but it’s not updated fast enough with all the latest techniques and scripts.

@Magnus919 3/ I also trained a model of myself with the DreamBooth technique via the @MysteryGuitarM Jupiter notebook and a @vast_ai instance. That was a lot of work too get the results you see in my pinned tweet.

Still trying to understand the extent of what’s possible on that front.

@Magnus919 4/ The hardest part is prompt engineering (at least if you want to create something original instead of regurgitating the same styles and outputs that thousands of others have tried).

I have created a rudimentary study generator that produces prompt files in particular ways.

@AndiMann Can’t reply to him as he has locked down the tweet

@Magnus919 1/ I spent weeks experimenting with Dalle-2 to generate original artworks in the style of famous painters I love. Some examples: https://t.co/gKozjGd7OR

Then, when @StabilityAI releases it, I started a deep dive in #stablediffusion

RT @OfirPress: We’ve found a new way to prompt language models that improves their ability to answer complex questions
Our Self-ask prompt…

It begins. >

Samsung Expects Earnings to Slump as Consumer Spending Slips https://t.co/8vxX8E7xls

RT @brezelradar: @moscioni_noah @dpa_intl @ELINTNews jup https://t.co/5XO4gG2Hha

RT @maxseddon: Ukraine’s postal service is already getting in on the trolling