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 Juggernaut, Wyvern Mix and Shonin Photoreal models to the mix to dilute down an issue with the eyes, added the EnvyGoodHands lora to try and fix the hands, and then used the Merge Block Weight tool to merge with v5 again to try and get the brightness and details up
FAQ
Comments (12)
Do I need to use the trigger word for version 6?
You can if you want to - I've had more realistic results without it, so have left it out this time around
would you be able to repost the image of the girl in the forest, I am a begginer and I wanted to test some things starting from it.
I havent deleted anything? Do you have it set to show everything??
The image may also be on one of the older versions?
that is weird:
"No showcase images available
No images from this creator match your content preferences. Adjust your settings or explore the community gallery below."
It was in V5.0 sorry
@jov141 No worries, I've found this model also had the issue where the auto-tagger didn't tag the images as I uploaded them which stopped them from being seen by people that weren't logged in. That's now fixed :)
Wow! Great results buddy... I am a beginner and also trying to get a checkpoint which shows photo realism... but am unable to figure out exactly what is the issue with my fine-tuning procedure... Can you give some step-by-step instructions, like how did you reach this checkpoint? I would really appreciate it.
Step by steps is a bit of a stretch, but your best bet is to get a1111/vlad if you dont have it already, then find a few of your favourite models on here and just try merging them using the standard "Checkpoint Merger" section with a "Weighted Sum" merge.
I keep most of the instructions in the version info for each version, but it isn't a full recipe as I don't remember what weightings I used through the process.
Once you get confident with merging with that tool, you can then start to use the Merge Block Weighted extension - https://github.com/bbc-mc/sdweb-merge-block-weighted-gui - which then allows for you to get more specific with the merge, and get better results.
Lots of trial and error, but when you get a good result, it's worth the effort :)
@residentchiefnz Thanks a lot man✌️✌️
OMG, i started using this last night and this has quickly become my favorite go-to realistic checkpoint. it's super sensitive to my prompts from other checkpoints, so i had to tone down a lot of things, but the capabilities of this checkpoint out of this world. easily one of the best checkpoints, period!
Indeed. It's impressive how you don't need a lot of words in your positive prompt. Love this model.