If you do like this work, consider buying me a coffee :)
Updated - Mid Year: Updated with a lora created from a frest dataset merged with ICBINP Final. Should give decent results, but does have the standard AI issues (cleft chin bias etc..)
Recommended settings: DPM++ 3M SDE Karras or DPM++ 2M Karras, 20-30 steps, 2.5-5 CFG (or use Dynamic Thresholding), happiest at 640x960 with a hires fix on top, but if you are happy to hunt through seeds to avoid head duplicates, 768x1152 works quite well also.
Following on from Gorilla With A Brick, I've merged in 10 more photorealistic models at various weights, and some more noise offset to create something that when prompted for photorealism will make you go "I Can't Believe It's Not Photography". It will happily create CGI characters and awesome landscapes as well.
As always, pruned to fp16, the VAE is baked in (SD-v2 840000 - ignore the VAE requirement above as the system is borked!), and the man in the suit can be used as the calibration prompt on any UI!
Description
Added in Edge of Realism, EpicRealism, HodgePodge and CyberRealistic to the mix, for more goodness!
FAQ
Comments (2)
I can't believe it's a photo!
Can you add comments to the grid comparison images to explain the differences that you perceive as improvements? I honestly don't see any version as the clear winner for photo-realism. While V2 & V3 are more frequent "winners", there are also some V1 wins in your examples IMO. I feel like all versions just become 2GB seeds to iterate results through.