>>14781

There ya go. ^^

Also here: https://files.catbox.moe/gsufkf.rar

Pw: gurochan ...content explained further down in this post.

>>14782

Oh dear...

Anon summons wall-of-text-mon... super effective!... >.< :D

I'm using Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge?tab=readme-ov-file ...though sadly it hasn't been updated in quite a while. If you have any gforce 10xx-40xx series GPU it should work just fine with the 1-click-installer, but if you have a 50xx series gpu you may have to do some manual tweaking and installing of a few dependencies.

There's a newer fork of Forge that is actively being maintained, but the downsides are that the installation is a bit more complex and from what I understand a bunch of previously built in extensions have been removed from that build. Not sure if "multi-diffusion" is one of the removed functions, haven't been abled to confirm either way, but multi-diffusion is essential to the "workflow" I use, so it is one of the two reasons that so far kept me from checking it out for myself. However from what I understand it supports Video generation and WAN, so if you're more tech savvy than I am and, unlike me, you understand the install instructions you might want to check it out: https://github.com/Haoming02/sd-webui-forge-classic/tree/neo

The base model I'm using is this here: https://civitai.com/models/997160/knk-luminai ...Also this VAE: https://civitai.com/models/140686/fix-fp16-errors-sdxl-lower-memory-use-sdxl-vae-fp16-fix-by-madebyollin (don't ask me what a VAE is exactly, all I know is you need one and this is less resource hungry than the default one/I'm not even sure if Forge comes with the default VAE)

...these are the Takabe Eri- and artstyle - Loras I used: https://civitai.com/models/2127793/takabe-eri https://civitai.com/models/1588942/illustrious-xl-18-in-1hidero-horibe-shuusaku-kisaku-artist-style ...the background - Lora (the creator has tons and tons of background loras, recommend you check them out)...: https://civitai.com/models/2041598/billionaire-bathroom-illu ...one of the guro-related loras I used can also be found on civit: https://civitai.com/models/667545/headless-style-artstyle-or-pdxl ...meant for pony and on its own it isn't working too well, but it helps to steer things in the right direction for decapitating stuff. I'm sure there are better loras out there by now that do this better but I haven't tried out any for a while... Then finally for the amputee stuff and general goryness I used the lora from this post: https://boards.guro.cx/ai/koko.php?res=3818&page=0#p13_3826 ...and: https://mega.nz/file/AFx2FKRL#aScqT2fnIPMjCeT1hkPd7e9Ha74KeAwQGGBLs0Rahrw


I've zipped all versions of the Takabe pic, also if you look close in the pic in this post here you'll see that the little blood-smudge on the edge of her mouth is gone; once again I had forgotten to re-enable the layer covering it up before posting the original. On the other hand it looks almost as if it was a thumb print - as if the killer had lovingly touched her mouth there. But her face looks a bit nicer without it imo. So since I couldn't decide which is better I included both version of her face both with all debreast(both/left/right)/pussystab versions in the zip file. Also I've included the raw initial gen for this pic;
once you have forge installed and running you can drag and drop the pic into the prompt input field and it will read the generation parameters (you just need to click the little blue arrow icon next to the prompt field after dragging the pic in to apply it)

You'll see the initial gen is pretty crude and shitty looking. The next step then was to make it sharper and better detailed by first upscaling it by 2 (in the "extra" tab in forge) and then sending that upscaled pic to the img2img tab and run "Multi-diffusion" on it. In the archive in the folder with the raw gen you'll also find the upscaled version of the raw gen with a bunch of preliminary manips I made by hand with paint.net (the photoshop-alternative I'm using: https://www.getpaint.net/ ). Flick between that second and the third pic in that folder and you can see the difference multi diffusion makes. From that onwards it was a matter of inpainting over almost every detail of the image and occasionally also photoshopping some further details to help/guide the AI until I was satisfied with the result.

The debreasting then again was done by first crudely photoshopping that in by hand and then running inpaint first over the stumps with the Lora from here: >>13980 ...and then over the severed breasts with the Lora from here: >>1909. I'll add an example for both these steps into the archive as well, same deal here, you can drag and drop it into the prompt field and you'll see the prompts. When you want to use them on other chars I'd replace the char- and artstyle - Loras of course...


I also still owe user Puskiller sort of a guide/explanation for this whole
"workflow" and I had started to do some sort of huge flowchart thingy to get all the relevant info into a better understandable format. Every time I try to explain it in a way that even a total newcomer can understand, I end up using so much "expert-lingo" ~ terms that my shitty old boomer brain takes for granted after working with this stuff for so long. And without a proper visualization I feel explaining via text alone won't really help. So that guide thing I started has already become a huge file (since so far I've been too dumb to make it in another format than just a 30kx30k pixel .png/jpg image.) and because of that I ran into a dead end with it so far. (The huge pic with tons of layers crashed my drivers, making my (newly built) pc behave as if it was dying for almost a minute with both monitors repeatedly turning on and off, sending me into yet another panic attack >.< xD)

I will not take up new requests for a while to get to finish/fulfill some other projects and requests for a bit now and #1 on that list will be to somehow get that guide/tutorial/"workflow" - explanation/demo together.

So stay tuned for that and maybe see if you can get forge to run and start experimenting in the meantime. :)