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 new versions of Avalon TRUVision, CyberRealistic, Huns, and MagicMix Sombre to the mix at a flat rate merge, added the Detail Tweaker Lora back in, then did a merge block weight mix with ICBINP_v6 to reduce some of the details
FAQ
Comments (25)
Really love these models great job. V7 seems worse then V6 for me. More chance of a borked face.
Thanks for the feedback :) Will look into it and see what I can do to make v8 better!
I have the same experience from preliminary tests.
@Brobb @argh Interesting.. do you have a particular prompt/neg prompt/settings that looks better on v6 than v7? v8 is in testing and I would like something that would confirm this will be fixed for you!
@residentchiefnz Sorry for late reply. This was bad on v6 and better on v7:
Emma Watson holding a wood handled axe, wearing transparent white raincoat, looking (desperate:0.9), (grimace:1.2), furrowed brow, (clenched teeth:1.2), in apartment, cinematic, Fujifilm XT5.
Negative prompt: bad-picture-chill-75v, badhandv4, bad_prompt_version2, Unspeakable-Horrors-Composition-4v, border, frame, text, watermark, signature, (out of frame:1.2) , jpeg artifacts, signature, watermark, username, blurry, artist name, render, 3d, 2d, sketch, painting, digital art, drawing, disfigured,
I haven't finished my tests of v6 and there is v7 now :-)
If I keep finding things that I think make the model better, I add em :)
@residentchiefnz i'm not complaining, the samples look really nice :)
Wow, I have been playing with the latest version and have been getting some beautifully creative results, lovely detail as well. A noticeable improvement, well done and thank you.
I'll give this a shot, but looking at the example photos and all the submissions, I notice zero pictures of women in their 30s-50s.
This model will only be truly good if it can produce female output in that range without pounding the crap out of the negative prompt box. Hot sauce or grandma is not only two choices I wish to have.
Good observation and fair call. Ill get a couple of x/y plots of age and gender with simple negative prompts up later today
@BigBlueWolf https://civitai.com/posts/237544 - a couple for your perusal and feedback
@residentchiefnz cool, how do you use that?
@Tremescent As in how did I make it or what is it?
To the latter, it is a prompt substitution so all inages have the same prompt, but the age and gender change so you can see how well or not well the model handles the age
@residentchiefnz In the first set (purple-ish backgrounds) I would guess her ages to be:
15, 20, 35, 45, 50, 55, 60, 65, 70, 75.
The guy starts out OK but realism really stats falling apart at the 35 mark.
Second set generally looks better but both male and female look increasingly unrealistic at above 45 mark. 20 & 25 female both look same age to me, 45 is odd. she looks like she's been a heavy drinker or something.
IMHO. Others may disagree.
Two examples sets aren't really enough tests to make a definitive conclusion, but I'd say in some cases, yes, the model skews younger than target age for women at first, then shoots past the mid 20s through early 40s.
@BigBlueWolf All good, something to test for when making v8
Models handles age perfectly,probably one of the best models around that really knows how to build a proper human character of any age,but you really need to prompt it in a right way.
I was using V6 in A1111 with no issues, V7 though will not load "Error while deserializing header: MetadataIncompleteBuffer"
I would delete and re-download, suspect something has corrupted/not fully completed
@residentchiefnz you were correct, this did indeed solve the issue, thank you
Yeah, v7 definitely works just fine in Automatic 1111.
download the config file too
@abrahamaraon Hi, where can I find the configuration file?
@The_Moye There isnt one…
@residentchiefnz Oh right, so that was a trick to make me lose my mind 🤪