>>1835320 Nope. >>1836369 Anon looks cool. Viv looks eager to swallow semen though.

>>1790296 What style did you use for the Doll?

Again my images get randomly deleted from Civitai for no apparent reason. Two people even gave me buzz but no, you can't have shit. >>1839118 masterpiece, high quality, detailed background, realistic, (intricate details:1.1), (kotteri, masha:1.2), (harada takehito:1.2), (depth of field:1.2), (plain_doll:1.2), 1girl, solo, doll joints, tucking hair, facing to the side, sideways glance, looking away, expressionless, (mature:1.2), portrait, skinny, eyes half closed, long eyelashes, colored eyelashes, grey eyelashes, bonnet, red ascot, brooch, shawl, black clothes, brown trim, pattern print clothes, white hair, grey eyes, copper roses, hat flower, dark background, The doll prompt I got from an Anon on 4chan not sure where, maybe it was on /g/. >>1836369 Those look clean.

>>1839979 >my images get randomly deleted from Civitai for no apparent reason >people even gave me buzz but no, you can't have shit After seeing what indians did to the payout program on X I've no doubt that civitai's system is being manipulated by malicious actors. And like X, they'll do absolutely nothing to deter exploitation of their reporting system.

>>1839979 ty (･ω･)b >harada takehito Nice. It looked very familiar.

This Medli image started with aliased lines, but partially anti-aliased them during inpainting for some reason. To try & remedy that I installed a pixelization extension on Forge-Neo, but it did not appear in Extras tab as claimed. So I tried installing a different Pixel extension & it bricked Forge-Neo. Tried downloading & replacing everything mentioned in the traceback error, but it wouldn't start anymore. So I had to reinstall Forge-Neo. Thankfully it copied some things locally since it didn't download Torch again which takes ages. I kept wheel files locally to install Sage Attention 2.2.0 & Flash Attn 2.8.2 so I didn't have to go find that shit again. I copied my configs & junk back to it so I didn't have to set things again. I dunno how one extension bricked it. Second image is some degeneracy for /loli/. Inspector Gadget video games exist so I'll post it here too. Very annoying pose to generate. LoRAs were linked there. >>>/loli/9863 >>1839979 That Phil Bourassa style generally makes clean, thin lines. >>1841301 Cute. Reminds me that I never beat Chrono Cross. I remember playing it, but IIRC it was a rental.

Orenjipiiru style: https://civitai.com/models/111792 I did NOT use a character LoRA, but later saw one: https://civitai.com/models/1030924 Maybe should've since the character LoRA would've got hair color right & I wouldn't have had to photoshop that.

>>1841808 Heh. I remember Queen Zeal kicked my ass so bad I didn't finish Chrono Trigger. The save is probably still there too.

>>1843579 Tried using character LoRA this time. Mixed Orenji style LoRA with the character, Minimalist Inkwash & Pixel Art LoRAs. Pixel Art IL: https://civitaiarchive.com/models/466454 Minimalist Inkwash IL: https://civitai.com/models/1534535/minimalist-inkwash-illustratious <lora:orenjipiiru_v6:0.8>, orenjipiiru, artist:orenji_\(orenjipiiru\), simplistic, toon_\(style\), flat color, sketch, (hatching_\(texture\):1.2), <lora:Pixel_Art:0.6>, pixel_art, <lora:minimalist_inkwash_il:0.6>, minimalist, ink wash, traditional media, (sketch, graphite \(medium\), watercolor \(medium\):1.2), partially colored, soft color, Increased weight of Pixel Art LoRA during inpainting to keep roughness of lines. Hair color still wasn't entirely correct & dress color was almost always too dark so still had to tweak in photoshop.

What, so we have two AI generats or something? I posted >>1844481 in the "wrong" thread? tl;dr I'm trying out Forge Couple to mostly abhorrent and barely successful results. I don't think I'm good enough with inpainting to fix the ones that kinda-sorta understood my prompts. Rechecked >>1717581 for prompt format guidance (assuming they work on Forge Couple) but as stated before, I suspect those who use these advanced tools are very rare.

>>1845040 >>1844938 Shouldn't have errors. What Stable Diffusion webui are you using? Forge? Forge-neo? A111? Maybe reinstall Forge Couple, or else try Regional Prompter instead. To uninstall, delete "sd-forge-couple" from the "extensions" folder in webui directory. Pic#4 here >>1717581 shows where in webui to install extensions. Extensions tab - Available tab - click Load From button - CTRL+F search page for name of extension - click Install button next to extension - close all Stable Diffusion windows - restart Stable Diffusion To reduce RNG use a guide image in Controlnet. This forces the position of characters so you can mask them accurately with Forge Couple. Mistoline: https://civitai.com/models/441432/mistoline goes in "models\ControlNet" directory of webui folder. Rename file to "lineart_mistoline_v10.safetensors" & it should appear when you select "Lineart" in the Controlnet section of webui. I made new screenshots showing Forge Couple & Mistoline usage. You can also use "Region Assignment - Advanced" for the inpainting in pic#4, but for me that's janky & can delete prompts so I just use masks. 0.8 weight - Sinozick Style LoRA: https://civitai.com/models/1324670 Saria LoRA: https://civitai.com/models/1888313 Tetra LoRA: https://civitai.com/models/546968 I accidentally used an old Pony LoRA for Tetra instead of an Illustrious LoRA. Lack of sleep.

>>1846460 Thank you very much! I'll certainly try this later today (with proper vidya lolis) but until then, I did manage to make the "get_mask" error go away. Maybe. Turns out Forge Couple (now at 4.0.4) was due for an update, among other extensions. The error message no longer come up but the results still weren't there but I suspected my prompt for the faceless ugly bastard was another issue. I switched to Advanced, downsized a line to what you see in pic 1 and suddenly, most of them are proper! Not 100% what I want (I was trying to replicate pic 2) but the base prompt is obeyed now. By the way, is the renaming of Mistoline required I see it shows up just fine in the Model drop down and the Preprocessor drop down seems to already have everything, including stuff like Openpose and Canny. Saves me a bunch of time scrambling to God knows where to download a bunch of stuff. Oh, and I use Forge. No "re" or "neo". Also later I'll try recreating my success with the mask method, if the prompt really was the problem. Do you make your own guide images or can I screencap some random JAV or snip away a panel from some h-manga and hope for the best? My skill withe a mouse isn't good but early experiments with inpainting turned my subpar scribbles into pretty neat things. Thanks again. Will report back when able!

https://civitai.com/models/1145434/mavis-dracula-hotel-transylvania-illustrious https://civitai.com/models/2043904/viper-series-oekaki-retro-style-illustriousxl-noob-lora >>1847463 Neat.

0.9 weight - Carol Onirism IL LoRA: https://civitai.com/models/1254421 0.6 weight - Orenjipiiru IL LoRA: https://civitai.com/models/111792 0.3 weight - Minimalist Inkwash IL LoRA: https://civitai.com/models/1534535

Zonos didn't get Miku's voice right (tried to make it sound more natural than it is) so I tried this: https://huggingface.co/spaces/John6666/mikuTTS Only the JP option sounds right, but has obvious issues with engrish. >>1846599 >is the renaming of Mistoline required Probably not. I just rename it so when I select Lineart option it narrows things down to what I need. If it's already doing that for you then leave it as it is. >Do you make your own guide images Yeah, I trace specific things I want from an existing image or fabricate something simple. Can also just paint out the background of a photo. See examples: >>1718446 >or can I screencap some random JAV or snip away a panel from some h-manga and hope for the best? You can use pretty much any image for Mistoline. Here's another example. The logo was not generated obviously,

>>1849889 I suspect those with a drawing tablet will have the biggest advantage. I have GIMP, which has something called the Path tool which, according to a video I very fortunatley bookmarked back when I was screwing around with inpainting, seems to be able to trace and make outlines of whatever. Again, no confidence with freestyling with a mouse so this may prove to be a Godsend if going without a guide image proves frustrating. Oh, and the promised vidya loli blowjob. Again, I'll follow that very helpful guideline above but I wanted to give style mixing a try. Or at least get my toes wet. I learned that some (like Bible Black) are impossible to gen lolis with, so I went with the likes of Bleedman and Atte Nanakusa at 0.4 weight. Can't say if I have a keen enough eye on whether or not pic 4 is a successful mix but Carrie's eyes won't stay red. https://civitai.com/models/125172 https://civitai.com/models/954137 https://civitai.com/models/241207?modelVersionId=1411407 https://civitaiarchive.com/models/1009715?modelVersionId=1131773

I tried to bake. Tell me what you think: https://civitai.com/models/2047167/neamosub >>1848650 Cute.

https://civitai.com/models/1104659/megaman-legendsrockman-dash-ps1-style-illustrious >>1854429 I prefer flat chest or small breasts Marie. Neat art style. >>1850288 Tried it. Couldn't get same results on JANKU. LoRA was on 1.1 weight, but I guess it doesn't behave the same way on this checkpoint. "super deformed" AKA "chibi" tag introduces extra little people a lot so I have to use another LoRA to prevent that. Was also getting a border although border was in negative.

>>1854793 Thanks. It's a mashup of a few styles. > LoRA was on 1.1 weight >Was also getting a border although border was in negative. I had this problem too. The training data was only 38 images so I just updated the trigger words to reflect how weak it is. The border problem is Janku though. Most of the training images are white_background, simple_background.

https://civitai.com/models/1340056/the-lord-of-the-rings-style-illustriousxl https://civitai.com/models/1371245/ninja-scroll-juubei-ninpucho <lora:the_lord_of_the_rings_style_illustriousXL_r1:0.7>, lotr_style, <lora:ninja_scroll_IL:0.4>, anime_screenshot, retro_artstyle,

>>1854966 It's kind of easy to get porn from Grok.

>>1859900 Using this style mix again. Creates a detailed retro anime screencap aesthetic. >>1861563 >Marie's sausage became a shoe >then it tried to give her a penis lol >kind of easy to get porn from Grok They trained Grok on Reddit posts which made it gay & retarded. Tried using WAN-Video in Forge-neo, but all it produced was a black PNG. I have multiple scripts that use ffmpeg & it's added to PATH so Forge-neo should've at least encoded a video, but it didn't. I still have ComfyUI set up for WAN, but far too often its outputs are not clean. Guess I'll stick to Framepack since at least I can see what it's making & cancel if it's trash. It's all very slow. The only "quick" video generation I tried was a WAN Self-Forcing DMD VACE setup, but there is less & less motion in the video the higher you set duration beyond 5 seconds due to limitations of the method which made it useless IMO.

>>1863991 >It's all very slow. What GPU?

>>1859900 Video gen for this. Framepack (Hunyuan) cannot retain the moving text, but can effectively bounce an ass around with jiggly-everything and jelly-hips LoRAs. Unfortunately it's moving her arms in stupid ways. Should've zoomed in to crop out problem areas as usual or else generate an image with an easier pose for it to not fuck up, >>1864451 RTX 5090. Local video gen is very taxing & speedups like TeaCache increase motion artifacts so I can't even enable them without further muddying the output. I looked to see if Framepack or Framepack eichi got any updates. Nothing. On Framepack git the dev posted YESTERDAY that they're still working on P1 https://github.com/lllyasviel/FramePack/issues/768 so that's proof-of-life at least.

I tried a LoRA for WAN 2.1 that let's you use LCM sampler at 4 - 8 steps for faster generation. A bit faster, but I still get the same visual issues with outputs that made me give up using WAN before. Going to try upgrading to a WAN 2.2 model. It has a lot of existing LoRAs Found a page that provides various checkpoint merges including one tailored for nsfw. https://huggingface.co/Phr00t/WAN2.2-14B-Rapid-AllInOne hopefully that means I can prompt nsfw slop without having to rely on LoRAs as much. Dunno if my ComfyUI install will support this junk with my existing workflows. I don't have a sampler they mention, but the version I'm getting uses samplers that I do have. Latest mega model is like 22.7GB so it's gonna take ages to download on my trash internet.

>>1865437 >>1865765 After a lot of frustrating effort I got some WAN 2.2 workflows "working" in RatNestUI ComfyUI. The WAN 2.2 VAE gives me errors so I'm forced to use 2.1 VAE. KSampler outputs black video if I use FP16, garbage with BF16, errors with FP32, so the only precision I can use that actually completes a video output is FP8. WAN 2.2 retains the art style & text better than Hunyuan (Framepack), but can give videos an over-compressed look due to some artifacts. I have another workflow using high & low noise models together which produces over-sharpened outputs.

>>1869298 The ass is fatter in this one.

>>1870245 Maybe due to the aspect ratio being stretched in the first one which messes with the roundness of it. Second video is 16:9 ratio like the original image so no cropping or stretching.

>>1854793 Some Vita idle animations. WAN-Video Model: Rapid WAN 2.2 I2V GGUF: https://civitai.com/models/1855105?modelVersionId=2182185 Used FJ LoRA at 0.4 weight to make her shift feet around a little: https://civitai.com/models/1364638 I like how WAN 2.2 will stick to the visual style rather than trying to morph into something else. I often had to use anime LoRAs with Hunyuan otherwise it'd start adding lips & softened shading of a "cel shaded" input image, but WAN 2.2 hasn't done that. Additionally, WAN 2.1 LoRAs appear to be compatible with WAN 2.2 so all my unused WAN LoRAs that were taking up drive space could be useful again.

>>1854429 >my reaction when Marie makes that face at me

>>1776374 Surprised it could follow the motion of her long ponytail, but it kept wanting to add footwear & setting a longer duration made visual artifacts.

>>1712471 I struggled to get Hunyuan (Framepack) to do this, but WAN 2.2 was finally able to. I used the innie pussy LoRA for WAN 2.2 & although the area is darkened a bit with a gradient it did work. It was nearly two months ago that I used a LoRA of the same name on Hunyuan (Framepack) without success & eventually resorted to keyframes which at best just slide them off without her even touching them. >>1715955

NEW THREAD: >>1875633

>>1875671 We did it! Let's go!