Post AhdsQMvQ4PfLSRZ28m by j@jaesharp.social
 (DIR) More posts by j@jaesharp.social
 (DIR) Post #AhdsQMvQ4PfLSRZ28m by j@jaesharp.social
       2024-02-08T21:20:35Z
       
       0 likes, 0 repeats
       
       Needless to say, photomaker guidance does way more than just human faces!This is base SDXL using Photomaker guidance (realistic) - https://huggingface.co/spaces/TencentARC/PhotoMaker - on anthropomorphic kangaroo portraits (No human faces used for guidance only anthro kangaroo portraits, neonpunk Style, ~30% Style Strength, Guidance scale ~8, Sample Steps 50, prompt "kangaroo man img" - total images generated to select this one was approximately 1k taking ~30 hours of A100 time with a non-scuffed gen rate of about 3%).
       
 (DIR) Post #AhdsQNqqctbSKYQtea by j@jaesharp.social
       2024-02-08T21:22:28Z
       
       0 likes, 0 repeats
       
       Most of the time, due to SDXL's base model not being tuned for it, and also the photomaker guidance model's human tuning, a lot of work in parameter tweaking/selection is required to get it to this kind of image - but with enough tweaking to the style percentage and guidance scale, the random seeds for images like this are in there... keep working at it until you get something that works as awesomely as this does! (especially for non-fine-tuned SDXL using a _human specialised_ guidance model!)
       
 (DIR) Post #AhdsQODBHrT3RosjSq by j@jaesharp.social
       2024-02-08T22:08:42Z
       
       0 likes, 0 repeats
       
       BTW, this is, of course, a demo what can be achieved with some brute force and a new guidance technique on top of a base model - and not of creating a piece shaped by any one particular vision I had. That requires far more care and work than this. For me, this piece is the equivalent of a doodle and isn't representative of the usual work and care I put into my pieces. Just a bit of fun :)