Photorealistic Lora.
Tested with ryanblend_v60 and dreamshaper_5BakedVae.
resolution: 512 x 512 or 512 x 768.
Enjoy,
-
description update:
This Lora doesn't want to offend anyone, much less influence people to do the same. This Lora is only an experimental solution to evaluate the correctness of Stable Diffusion in generating images. Thanks for understanding!
Description
first experimental version, "Promt" advice is welcome
FAQ
Comments (9)
Personally doesn't phase me but I know this is going to be deleted any moment lol.
toddlers when you leave the gun safe unlocked for 5 minutes:
Wonder if it works with anime models
At least there are boobs. ♥️
Why do Americans lick guns?
25 MPixel Reconstruction did someone gone insane here ;)
But dammit it looks better then when Photoshop Igor would smuch over it with his frequency skills. :D
But we have to heavily argue about this results usefulness ;)
i would say 8-11 Mpixel and we should be happy as hell.
I really wonder how long the Civitai admins gonna let this bitrate be allowed and take it down eventually.
i don't think you can justify the 44-50 MB here for the overall Artificial result and total perceptual instability.
Any real camera Photo will kill this with way lower bitrate in terms of details (stability).
But stil it is a crazy interpolation result and thank you for showcasing these results lets see if we can improve them way further on the detail layer first of all getting rid of all that makeup layers :)
As long as that gun looks like a plastic toy it's a failure as well, some results are also totally frequency over sharped there is no new data at all just artifacts. which pump up the bitrate perceptual unnecessary.
But these results look way better than what any Anime Waifu Model would do and we see way to much of these abominations.
You should really take some look at some total natural humans and understand when bitrate is necessary and when it isn't.
Your results make me wonder what Codeformer would do if it tries to Codeform something which allready looks like 80% of it's trained beautyfication data bias.
Even results from very cheap cameras by now off yesterday can look way better when not to heavy artificialized.
no prompts ..clearly bad at sampling
Only stupid people make 40 MB preview images...
have you made a 2nd version of this?
Or other LORA yo uhave no uploaded here?