Finetuned the new SD1.5 model with some realistic portrait photos, ran a quick test of quality.
No GFPGAN/CodeFormer used, no inpainting - there's details I could Photoshop here and there, but ... Not bad. Quickly looking, one might be fooled.
#stablediffusion #portrait
No GFPGAN/CodeFormer used, no inpainting - there's details I could Photoshop here and there, but ... Not bad. Quickly looking, one might be fooled.
#stablediffusion #portrait
Few more from the same test.
Same prompt (minus token word change to use the custom model), same seed, same settings. SD 1.4, SD 1.5, finetuned SD 1.5 model. No Topaz.
There's a nice leap up from 1.4 to 1.5.
There's a nice leap up from 1.4 to 1.5.
"model eating ice cream" still needs a tad more work
oh god yes these are not cursed at all!
Continued here
https://twitter.com/rainisto/status/1583530257411297281
• • •
Missing some Tweet in this thread? You can try to
force a refresh
