[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/jp/ - Otaku Culture

Name
Spoiler?[]
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File[]
  • Please read the Rules and FAQ before posting.
  • [sjis] tags are available. Install the Mona font to view SJIS art properly.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


Janitor applications are now being accepted. Apply here.


[Advertise on 4chan]


File: 8766874876.png (1003 KB, 816x1264)
1003 KB
1003 KB PNG
Previous: >>48114121

WhAIterock Edition.

• A space for sharing and discussing Touhou-related AI content, namely RP chatbots, lorebooks and images.

• Chatbot Resources:
Frontends for chatbots:
>https://docs.sillytavern.app
>https://risuai.net
>https://agnai.chat | https://rentry.org/agnai_guides
Koboldcpp (UI for local chatbots):
>https://github.com/LostRuins/koboldcpp/releases/tag/v1.76
Example website with several touhou chatbots. Also allows you to create your own:
>https://chub.ai/
Akyuu Knowledge Lorebook (useful entry-level lorebook for giving your 2hu AI some context to RP with):
>https://chub.ai/lorebooks/hebelover/Touhou-AkyuuKnowledge
Botmaking guide and aid
>https://rentry.org/meta_botmaking_list
>https://desune.moe/aichared
>https://docs.chub.ai/docs/the-basics/character-creation

• Image Resources:
Advanced UI for local generation:
>https://github.com/lllyasviel/stable-diffusion-webui-forge
>https://github.com/Panchovix/stable-diffusion-webui-reForge
>https://github.com/automatic1111/stable-diffusion-webui
>https://github.com/comfyanonymous/ComfyUI
>https://github.com/invoke-ai/InvokeAI
>https://github.com/vladmandic/automatic
>https://github.com/mcmonkeyprojects/SwarmUI
Beginner site to browse for image models:
>https://civitai.com
Current 'meta' base image models:
>PonyXL
>illustriousXL
Index of guides for local generation:
>https://rentry.org/sdg-link
>https://rentry.org/rentrysd

• Requests are fine, but please avoid making the entire thread about them.
>>
File: 6532814569.png (777 KB, 1024x1024)
777 KB
777 KB PNG
>>
>>48172147
More boy wriggle/girl wriggle.
>>
>>48172147
more boyhus in general
>>
>>48172147
NoobAI is free and better than any paid service at this point. What's the point of requests?
>>
>>48172195
people are too dumb/lazy to set it up
>>
>>48172121
>Current 'meta' base image models:
>>PonyXL
>>illustriousXL
Please change this in the next bake.
>>
>>48172195
>local models
It's over
>>
>>48172195
tell that to my 980 ti
>>
File: 00018-2303239514.png (1005 KB, 840x1280)
1005 KB
1005 KB PNG
>>48172195
My best guess is not everyone has a strong enough PC to run local models
>>
>>48172578
Civit lets you run them remotely.
>>
>>48172578
>>48172616
Civitai is great for beginners or just fucking around. You get 500 (If I recall right) credits after signing up and you can spend them to gen Pony.
>>
>>48172616
Image generation models? For free? Without any restrictions?
>>
>>48172578
>strong enough PC
Kind reminder that, since reForge's optimisations, it is possible to run an SDXL model (i.e. Illustrious/NoobAI) with 4GB of VRAM.
>>
>>48172651
Wouldn't it be extremely slow though?
>>
>>48172147
Well endowed and healthy reimu and sanae in skimpy clothing casually sharing pamphlets in the town to gain followers.
>>
>>48172616
Interesting. What's the catch?
>>
>>48172947
you have to use CivitAI's credits system to generate images and you'll run out pretty fast
>>
File: 00029-2130259858.png (985 KB, 824x1176)
985 KB
985 KB PNG
So, having switched from the latest version of Forge to reForge, I haven't really noticed any changes so far (in terms of performance and results) but maybe I'm either just faiing to do so or I haven't set it up properly yet.

Are there any extensions I should get/disable? I already got booru autocomplete.

Also, I was told my settings should supposedly be saved but that doesn't seem the case. Am I just being stupid? To be fair I don't actually know if I have to save it manually or anything like that, I can't find the option to, at least. Whenever I'm done using reForge I just close the browser window and command window.
>>
>>48173263
>haven't really noticed any changes so far (in terms of performance and results)
Results are supposed to be the same. Performance may or may not be better depending on your hardware (bigger change on slower setups). If you're on an Nvidia card, you might want to put --xformers in command line arguments.
>Are there any extensions I should get/disable?
You can disable every extension starting with "sd_forge_", except sd_forge_neveroom. And maybe ControlNet, if you plan on using that.
>I was told my settings should supposedly be saved
Settings are saved in ui-config.json. If you have another Forge installation, you can copy the file over. Otherwise, on a clean webui launch:
- don't enter a prompt
- set everything (sampler, steps, resolution, what have you) how you want it to be by default
- go to the Settings tab -> Other -> Defaults, then hit View Changes and lastly Apply
This'll save your settings to ui-config.json and auto-set them to the specified values on next launch.
>>
>>48173263
>Are there any extensions I should get/disable?
Images Browser, adetailer, and stealth png info are invaluable.
>>
>>48165223
NTA but here

https://litter.catbox.moe/or1qzl.png
>>
>>48172578
holy milk truck
>>
File: 00005-516023858.png (603 KB, 800x1296)
603 KB
603 KB PNG
>>48173288
>Results are supposed to be the same.
Are they though? I thought they'd be different because of, supposedly, less handholding, no longer secretly modifying the values sent to the model, etc...
>You can disable every extension starting with "sd_forge_"
Got it, thanks.
>And maybe ControlNet, if you plan on using that.
Maybe. What is it?
>Otherwise, on a clean webui launch:
Thanks! Helps a ton. I also had another question come to mind, where do you guys save your prompts, if you do? I've been copy-pasting them on a text document, it's been working so far but it feels just archaic enough to give me the feeling that I'm not doing this optimally... haha.

>>48173299
>Images Browser, adetailer, and stealth png info are invaluable.
Thank you! I'll install these as well.
>>
>>48172578
I can feel my pelvis shattering just from looking at this image
>>
File: 05190-3403804283.jpg (2.46 MB, 1400x1800)
2.46 MB
2.46 MB JPG
Ran, you idiot, do you need a recalibration? There's no way that this is the right ga-

>>48174743
The problem with Forge is that it changes your settings to hard-baked values (that the devs deemed "optimal") from script files when you click seemingly unrelated elements of the UI, and some of those changes aren't readily apparent until you examine the metadata closely. If you load the same model, prompt, seed, CFG and so on (basically, the same metadata) from a pre-existing PNG into reForge, the result is going to be the same - provided your RNG settings are also the same. They should only be different if you change K-diffuser, GPU/CPU/NV RNG settings or run it on a different graphics card.
>What is it?
ControlNet or CN is a semi-advanced composition tool. It lets you, among other things, generate characters in specific poses extracted from other pictures, generate backgrounds from rough sketches and other nifty stuff that 95% of the userbase will never find a use for. There's a myriad tutorials on YT if you want to get into it.
>where do you guys save your prompts
They're saved by default inside your output files. You can view those from the PNG Info tab. There's also buttons for copying them over to t2i/i2i. For regularly reused parts of the prompt (quality positives, negatives, styles), you can use the Styles gizmo underneath the Generate button. You enter the string, save, and next time around just pick the name of the "style" from the drop-down list, and it'll append all the tags contained therein to the end of the prompt, without you entering them manually.
>>
>>48174811
DEATH BY SNU SNU
>>
>>48172578
doesnt look that impressive
>>
File: 00009-3486405540.png (842 KB, 800x1000)
842 KB
842 KB PNG
>>48174829
>There's a myriad tutorials on YT if you want to get into it
Sounds like the type of thing I'd use after 3 hours of trying to generate a very specific image and having the AI gacha fail me over and over again, I'll look into it when that happens.

>>48174829
>They're saved by default inside your output files
Got it, thanks.

Cute hag, by the way. Big fan of clumsy women.
>>
>>48172147
Kasen reverse raping you while only wearing her hat and tabard
>>
>>48173299
I was able to get the other two, but I can't find stealth PNG info for some reason.

Also, it feels like reForge is a lot more demanding on my PC for some reason. Are there any extensions to reduce how much VRAM it uses? Preferably without affecting generation speed too much, but I'm willing to make the sacrifice if I really have to.
>>
>>48175144
>Kasen
>her hat

What?
>>
File: 00053-2205350314.png (1.25 MB, 816x1224)
1.25 MB
1.25 MB PNG
What are some of your go-to artists for image generation?

For me, it's Hisona. Their style (and me wanting to see more of it) was the main reason why I gave AI art a try. I was really happy to see that, aside from making the eyes look a little derpy sometimes, AI can replicate it pretty much flawlessly.
>>
File: nue chang.png (1.76 MB, 1088x1920)
1.76 MB
1.76 MB PNG
>>
>>48172147
催眠 earrings on anyone except for Kokoro and Kasen
>>
>>48176889
Post an example, there's a lot of types according to Google
>>
>>48176015
heripantomorrow
kankan33333
kame_(kamepan44231)
wkar
spacezin
krekkov
naiga
>>
>>48183459
massakasama
>>
>>48181066
The big red earrings with the kanji on them, often worn by Kokoro and Kasen.
>>
>>48175284
Just use an eagle as a hat, problem solved.
>>
>>48172121
Why are you still spamming threads in /vg/?
>>>/vg/501621878
>>>/vg/501625779
>>
>>48184090
not OP but do you have any proof other than your singular brain cell to say that he's the same OP as the /vg/?

otherwise stop being a faggot.
>>
>>48184153
I was referring to the thread in general, stop raiding and spamming the /vg/ thread with avatarfags.
>>
>>48184090
>>48184167
meds now
>>
>>48184090
fillyfucker is a touhoufag?
>>
>>48184090
>Everyone I don't like on the internet is the same person!
>>
File: ComfyUI_01655_.png (928 KB, 832x1216)
928 KB
928 KB PNG
>>
>>48184090
You missed /g/ by, like, half the list. Check your aim.
>>
>>48184153
>not OP but do you have any proof other than your singular brain cell to say that he's the same OP as the /vg/?
The proof is that he's right here: >>48184405
>>
>>48184534
Why does his post have a ComfyUI filename instead of the randomized filename of the /vg/ thread indicative of someone using a ban-evasion service?
>>
>>48184548
Because he turned the option off to post in this thread.
>>
>>48184548
in fairness who else has such an autistically singular interest like koishi choking ryona
>>
>>48184090
Just because someone posted 2hu image or claim that they are from here doesen't necessarily mean that they are, more likely just crossposters flinging shit and you are just taking a bait, or maybe crossposter flinging shit is you.
>>
File: ComfyUI_01662_.png (1.07 MB, 832x1216)
1.07 MB
1.07 MB PNG
Don't reply to him please.
>>
File: 00001-2627931840.png (1.05 MB, 832x1304)
1.05 MB
1.05 MB PNG
>>48176015
kuraki
moriforest1040
258n
kumiyabe
choujiroo
rank_alee
the_atko
>>
Do any of you post your aislop anywhere other than 4chan? I need some vapid validation and thought about posting on pixiv or twitter.
>>
>>48184576
he is the crossposter flinging shit btw, /aids/ knows him by a bunch of different names, he started out there being a very angry anti-NAI schizo and then started frequenting pretty much every AI general on the site to stir up shit.
>>
>>48184778
He also frequents a group called 'oishicord, they organize most of the raids there.
>>
>>48184778
is he that shit-stirrer anon that other anon mentioned a few generals back about the "curse" of AI generals?
>>
>>48184742
One of the anons here posts on pixiv and to say he gets validation would be a bit of an understatement, he's really good though.
>>
>>48172578
birds shouldn't be this big
>>
>>48184405
>>48184604
What is this shit
>>
File: 00017-4092886621.png (1.07 MB, 840x1280)
1.07 MB
1.07 MB PNG
>>48184907
she's a grounded bird
>>
>>48184907
I'll have you know giant breasts are surprisingly aerodynamic
>>
File: 7511481748.png (83 KB, 235x352)
83 KB
83 KB PNG
>>48184834
It's just a small chunk of true autism and schizophrenia, not even kidding.
>>
>>48184742
Uploading to Pixiv sucks!
I'm sick of having to censor the dicks and clits I painstakingly redraw!
I'm sick of getting mass reported because someone has a vendetta against AI!
I'm sick of having to compress my pictures to shit because the upload page craps out if I try uploading anything over 2MB!
I'm sick of bot comments! Especially when they post suspect links I have a responsibility to manually delete!
I'm sick of viewers randomly asking for unrelated characters under months-old uploads! I'm not going to do Kaguya! Kaguya sucks! (Mokou gang wuz heer.)
No, I don't have a Patreon or a ko-fi! Stop asking! It's awkward, turning down well-meaning people's money! Makes me feel bad, like I'm spurning their generosity!
I'm tired of translating every caption and comment to JP! I don't even speak Moon! Thank you, Pedy! I now know the difference between the terms for a female and male virgin, but I can't exactly reupload the picture with fixed captions at this point! Can I!
And I don't know how Twitter works! Why on Earth was my account flagged for sensitive content TWICE, and what does that even do? Is it doubly removed from people's feeds now? In the first place, I'd already flagged it as such myself! So it's technically TRIPLE flagged! What!
>>
>>48185363
>No, I don't have a Patreon or a ko-fi!
If you don't want to get paid for indulging in your own hobby, that's your loss.
>>
>>48185363
How about creating an account on catbox and creating your album there?
>>
>>48185406
Fuck off grifter
>>
>>48185655
Does that mean you do it all for free?
>>
File: Uwah.gif (244 KB, 200x200)
244 KB
244 KB GIF
>>48185460
I didn't know you could make accounts on catbox, neat.
>>
>>48184834
Yes. He gets really mad about Touhou for some reason and a lot of the regular posters on /vg/aicg/ are Touhou fans so he tries to false flag them by spamming Koishi themed threads. It's kinda funny because I don't think he's self aware enough to realize Koishi's gimmick.
>>
>>48185757
Wait, you think it's Claudefag doing this?
>>
>>48185757
*Touhou avatarfags
And what about the little fact that the theme was picked in a fair way by random chance with a lottery???
>>
>>48185805
Yeah I think /aids/ calls him Claudefag sometimes. It's all the same anti-NAI schizo lurking every AI thread 24/7.
>>
>>48185363
How about Rule 34? It's a bit more degenerate, but at least you dont have to worry about censoring
>>
File: 1721648131835009.png (808 KB, 1024x1024)
808 KB
808 KB PNG
>>48172195
Noob, ironically, takes more effort to get looking remotely decent and its 1.0 release model is even worse than 0.75. PDXL and its derivatives Just Work despite being less flexible but its easier for people to get started on their (1girl,solo) first dive.

>>48185363
Maybe stop shitting up pixiv then you fucking sperg. You're not welcome there.
>>
>>48186760
What does that effort entail specifically?
>>
>>48187732
You have to load it up with every single jeet cope tag, artist, style, etc and the same for negatives before you even start prompting the shit you want otherwise you get incomprehensible blobs. PDXL you can unironically just type 1girl with nothing else or any negatives and get a normal looking image. noob 1.0 especially has a lot of edge bleeding from my uses and looking at civit galleries.
That alone sort of disqualifies the entire point of "noobai" being their self advertised "newbie friendly model" just on comparison.

Is the upper end of quality better? Subjective, but i've seen some really nice gens and much better flexibility out of noob 0.75, especially in regards to not needing loras for artists and styles out of the box. It'll definitely improve but again this is just focusing on their claims of being easier than other models.
>>
>>48185969
pretty sure you'd still have to worry if you're genning anyone who maybe veers slightly toward the younger side
>>
I've managed to get a comfyui workflow running to make pretty good Touhous, but is there any good way to upscale them, detail and clean up the images? the workflows I see on civitai are beyond me.
examples are catboxed, all advice welcome.

https://files.catbox.moe/7tgx4y.png
https://files.catbox.moe/7x430b.png
https://files.catbox.moe/5z7u31.png
https://files.catbox.moe/7674hd.png
>>
>>48187835
Aside from the weird resolution, these are pretty cute. Are these your first images?

I can't offer any advice I'm afraid, I'm kind of new to all of this as well and don't use comfyui. If it has img2img like Forge though maybe you could try upscaling the images there?
>>
File: 1697413015417327.jpg (194 KB, 1024x1024)
194 KB
194 KB JPG
Perennial request for any entertaining AI bloopers involving the Touhous. I love this stuff for some reason
>>
>>48187835
I went to download that LoRA, check the archives, and turns out it's mine, lmao.
anyway, for starters that isn't how you should upscale. there's a multitude of ways to do this but I went the simplest route which is a second sampling pass (img2img). also for sharpening you should try using self-attention guidance, since that works at the inference level.
https://files.catbox.moe/3h6s0v.png
if you want to fix the face you need to look into using facedetailer or inpaint-cropandstitch (the latter is good to use in that in teaches you what facedetailer is actually doing).
>>
>>48187875
I've only been fiddling with it for an afternoon or two, but using a NAI model and LoRA both posted here earlier. i spent seven gens fiddling with the ksampler settings and prompts to get things recognizably 2hu and my first properly good one was #34. the last box was #37 and is very similar. I'll try to give img2img another go, thanks!
>>48187923
If that's your LoRA, it's really fantastic! though it doesn't do raiko as well as it could. thanks for the recommendations, i'll post back with more as i learn over the next few days.
>>
File: catbox_4e07wp.png (766 KB, 1024x1024)
766 KB
766 KB PNG
>>48188081
>it doesn't do raiko as well as it could
not the lora's fault. try using related tags for the character, that should help it. https://danbooru.donmai.us/related_tag?commit=Search&search[category]=General&search[query]=horikawa_raiko
I'm not as intrinsically informed on how they should look but this seems pretty decent to me (didn't bother to clean this up). obviously more popular 2hus are gonna look better though.
>>
>>48172147
Kokoro pink bra paizuri
>>
>>48188143
maybe i'm using the wrong artists as influence, and i'll admit i wasn't tagging as much as I should when i was genning her. thanks for the input.
>>
File: 00026-905298912.png (864 KB, 896x1296)
864 KB
864 KB PNG
I've been messing around with implementing LoRAs into my gens for the first time and and think I've managed to finally come to a style that I really like. As much as I like it though, part of me feels a bit guilty because it's essentially just another artists' style (albeit one I really like) with just a bunch of LoRAs over it, hence my question:

I wanted to come up with a style that I could call my own and I wanted to do this by mixing different artists' styles (ironic, I know), so what's the best way of doing this? And for example, if I decide to use artist A and artist B as prompts, how do I control how much influence over the 2hu each artist has, as well as which parts they affect? (contours, coloring, shading, face, proportion, etc...)
>>
>>48188081
Why did you pick Comfy? I admit I'm new as well, but so far I've been finding Forge very accessible.
>>
>>48188541
I tried the easyai from the OP rentry links, didn't like it and moved on to comfy since it was the next down that would work on my machine. haven't tried reforge but i will now.
>>
>>48188502
can you take her clothes off?
>>
>>48188851
Can you take YOUR clothes off?
>>
>>48188886
I'm not doing it until she does
>>
File: 9208007979.png (1.36 MB, 832x1216)
1.36 MB
1.36 MB PNG
>>48188886
It's gonna cost you.
>>
>>48188502
>if I decide to use artist A and artist B as prompts, how do I control how much influence over the 2hu each artist has
if you're using loras you just weight them as you see fit. if you're using built-in model knowledge via the prompt, compel weighting is the only way to really do this, but only comfy supports it currently. I'd like to try porting it but the webui's text encoder code is a fucking hellscape.
>as well as which parts they affect? (contours, coloring, shading, face, proportion, etc...)
the face you can style individually via inpainting, the rest you're at the mercy of how much you mix them really...
it's possible to apply different lora strength values at different portions of inference, but this is still only something a1111 supports to my knowledge. not even re/forge or comfy (though comfy is working on it). for built-in model knowledge this isn't a problem and you can use prompt editing -- which is super powerful and unfortunate not many use it.
>>
>>48188903
Okay, not the original anon who gen'd the image but here

https://litter.catbox.moe/a0ml5x.png
>>
>>48188926
What is compel weighting? How is it different from usual?
>>
prince
https://files.catbox.moe/qffxe3.jpg
blubun
https://files.catbox.moe/brur40.jpg
riguru
https://files.catbox.moe/adrafn.jpg
>>
>>48188926
>If you're using Loras you just weigh them as you see fit
I guess I could try finding a few more style-related LoRAs in that case. They're kind of similar to an artist tag, in a way.
>compel weighting is the only way
That's a shame, I thought you'd be able to do it the same way as LoRAs (putting it between brackets and attributing a strength value to it). Part of me wants to ask what compel weighting even is but then again if I can't do it there's no point in making you waste time explaining it.
>the rest you're at the mercy of how much you mix them really...
Double shame.
>but this is still only something a1111 supports to my knowledge
Not true! ReForge supports it, at least. I was messing around with it quite a bit actually when I was trying to get a good mix here >>48188502 .
>you can use prompt editing
Interesting, what is it, and is it difficult? Or is that just another way of describing processes like ControlNet and inpainting?

>>48188965
I'm the original anon. If you don't mind, could you explain how you did this? As in, screenshot the area which you've inpainted, together with the prompt and img2img settings (assuming you used Forge/reForge) and post it? I still haven't gotten the hang of inpainting and I suspect it's just something really dumb that I'm doing wrong.
>>
>>48172147
Marisa being held down by Reimu and Sanae while Alice shoves a broomstick up her asshole please sir.
>>
File: 3464357354687.gif (1.23 MB, 546x408)
1.23 MB
1.23 MB GIF
>>48193393

>>48120455
>>
Twice, apparently.
>>
>>48190147
I should've clarified it's only a problem with downweighting, but see https://github.com/BlenderNeko/ComfyUI_ADV_CLIP_emb?tab=readme-ov-file#weight_interpretation (check the dropdown for a full explanation)
>>48191758
>>but this is still only something a1111 supports to my knowledge
>Not true! ReForge supports it, at least.
you must be mistakening this with something else, it definitely does not. it's been an open feature request for almost a year now.
https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/68
https://github.com/Panchovix/stable-diffusion-webui-reForge/issues/36
as for prompt editing, you can read more about that here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/features#prompt-editing
>>
How many posts on danbooru does an artist need to have in order to be recognised by Noob? Or does it work differently?
>>
>>48194683
I've gotten by with as little as 50 but it's pretty situational. Just try for yourself. For artists who've been renamed in the last year you may need to use a different tag.
>>
>>48190532
Not too sure about that kind of confidence on Seiran although that pride is a moonie trait
I prefer my Seiran desperate but to each their own
I do like the Miko and Wriggle
>>
>>48195157
I also always imagined Seiran to not have any hair down there, moon purity and all, though I guess you could also make the argument for a bush representing purity
>>
upscaler is running, next to work on inpainting and face detailer

https://files.catbox.moe/2t49sl.png
>>
>>48172187
seconded but make them have sex
>>
>>48174743
This Alice looks predatory, I had better notify Kotohime.
>>
>>48196587
>houtengeki
Good choice, remember that you also have waifu2x.net if you want to upscale stuff.
>>
>>48172147
kinda forgot about it but here's koishi from last thread
https://litter.catbox.moe/9o0oox.png
>>
>>48198255
>waifu2x
did you just wake up from a coma, anon?
>>
>>48197456
there is already a picture of a boyhu having sex in this thread and you didn't even notice
>>
>>48198563
she looks cute
>>
File: 05286-1283408363.jpg (1.2 MB, 1400x1800)
1.2 MB
1.2 MB JPG
Sometimes, when concentrating very hard on her next article while sitting at Geidontei, Ayaya forgets she's not at home and starts fidgeting. You can glimpse some incredible things in those moments.

>>48195157
I, too, believed in a nervous wreck Seiran, once. Then TH19 rolled around.
>>
>>48200971
you can still do a nervous wreck Seiran if you set it closer to 15 than 19
since her only appearance in that period is AFiEU, you can have her at any point in that development.
does Aya sometimes forget that she's a person and start perching on her chair like a bird? does she sleep like that too?
>>
ZANKUROHUS1112211!!!!!
>>
File: 1700894372579714.png (93 KB, 786x1200)
93 KB
93 KB PNG
>>48200971
I don't know, she still cries when she arrives in hell and gets comforted by Saki. She's multi faceted and that dialogue with Saki shows it best. I think her confidence's a front that falls apart easily
>>
>>48201116
>she still cries when she arrives in hell
Anybody would! Well, anybody with more than two motes of grey matter under their dome!
>>
File: 1706575712196992.png (2.74 MB, 3500x4000)
2.74 MB
2.74 MB PNG
>>48201152
My cute little Aunn chan doesnt cry!
>>
>>48200971
Doesn't 19 let us know that Seiran feels lonely on earth and is unable to hold conversations?
>>
>>48201181
>anybody with more than two motes of grey matter under their dome!
I think Aunn is exempt under this rule. Her skull is vacant.
>>
File: ComfyUI_01737_.png (1.66 MB, 832x1216)
1.66 MB
1.66 MB PNG
heun and heunpp2 work really well with noob, given that the artist tag is strong enough.
>>
>>48191758
>>
>>48202679
Is "guidance" the NAI parlance for denoise strength?
>>
>>48202696
guidance = CFG (classifier-free guidance)
because NAI trained a true inpainting model there is no denoise strength (or think of it as always being equal to 1, but it actually works due to having been conditioned on inpainting)
>>
File: 00003-879414850.png (731 KB, 800x1296)
731 KB
731 KB PNG
>>48202679
Okay, I'm doing the prompts right which is what I was worried about, but I really don't get it in that case. For this image for example I'm just trying to remove this second hairband on Alice's head, so I selected it, added hairband to the negative prompts and used the settings in the catbox links, yet after numerous tries it either completely ignores that I want the hairband gone or it just adds weird smudge to the area I've inpainted.

Prompts: https://files.catbox.moe/w6qveu.png
Settings: https://files.catbox.moe/bvfabn.png
Result: https://files.catbox.moe/twqa7c.png
>>
>>48203023
Select more of Alice's head, not just the hairband. Give the AI more context to work with, and it will identify the hairband and remove it. Try it and show me the results.
>>
>>48203067
Will do, thank you, but in that case should I select every part of Alice's head except for the hairband that I want to keep, or just her entire head?
>>
>>48203078
Select every part except for the hairband that you want to keep is what I would do.
>>
File: yuyuko.png (2.36 MB, 1472x1472)
2.36 MB
2.36 MB PNG
>>48201060
have a yuuy
>>
File: 00013-1625624709.png (742 KB, 800x1296)
742 KB
742 KB PNG
>>48203084
It still took a few tries and there were still some times where it ignored the fact that I wanted the hairband gone, but at least it worked (even though when it did it'd occasionally be kinda wonky near the headband as you can see). Thank you!

Some of the other faulty results if you're curious:
https://files.catbox.moe/1y1j37.png
https://files.catbox.moe/7qm2w4.png
https://files.catbox.moe/6qug4f.png
https://files.catbox.moe/3w7sfs.png
https://files.catbox.moe/bnthfj.png
https://files.catbox.moe/eimrju.png

Is it normal to get this many mistakes or is there something I should've done differently?
>>
Some stuff I've already posted elsewhere, but eh. SFW-ish, this time.
Elly: https://files.catbox.moe/iw6ryk.jpg
Yam: https://files.catbox.moe/r0n2r4.jpg

>>48203229
It would've been much, much easier if you'd just opened the source picture up in whichever image editing program is your preference, smeared over the headband with a soft brush then fed it back to i2i and inpainted. Or, better yet, upscaled. More work, technically, but quicker and easier.
>>
File: Kuya!!!.png (1.29 MB, 898x898)
1.29 MB
1.29 MB PNG
I love my maidwife
>>
>>48203430
Again, I like your SFW stuff, glad to see you do it more often. I really like that Elly.

>It would've been much, much easier if...
Ah, you're right, I remember you mentioning that a few threads ago but I've yet to try it...

>Or better yet, upscaled.
Right - upscaling gives the model more pixels to work with, therefore refining the picture, etc etc, I remember that now too. Forgot about that as well...

How much do you usually upscale your images by? I know the actual resolution depends on the context of the image, of course, but do you upscale it by a certain % maybe?
>>
>>48204029
Not him, but I've been taught that a good idea would be to do upscaling in increments.

upscale by 1.25x, redraw to keep the idea in line, throw it back into img2img, upscale, repeat.

It keeps the AI from straying too far from the original image on a big upscale jump.
>>
>>48188188
https://files.catbox.moe/gezhru.png
https://files.catbox.moe/aayo9q.png
https://files.catbox.moe/owz0ae.png
https://files.catbox.moe/sm8knv.png
https://files.catbox.moe/p8vfd1.png
>>
>>48172121
I want Letty breastfeeding me.
>>
File: notbenben.png (1.16 MB, 1400x1000)
1.16 MB
1.16 MB PNG
>>48201060
https://files.catbox.moe/je4kg7.png
https://files.catbox.moe/7rp6ep.png
https://files.catbox.moe/jzu2yh.png
https://files.catbox.moe/l84z6y.png
https://files.catbox.moe/qnw6hb.png
https://files.catbox.moe/gqyzde.png
https://files.catbox.moe/g9h8fj.png
https://files.catbox.moe/xsu66w.png
>>
File: 04535-2771790513.jpg (676 KB, 1449x969)
676 KB
676 KB JPG
>>48204029
>How much do you usually upscale your images by?
x1.25 with SDXL. How many times depends on the end quality you're satisfied with. Me, I've cut my process down to two upscales (896x1152px -> x1.25 -> 1120x1440px -> x1.25 -> 1400x1800), but I used to do three with Pony and four with SD1.5.

The optimal denoise value shifts with the exact model, but I've found Illustrious/Noob derivatives swing a lot where Pony was moderately stable (0.4-0.5 range). The truly stable value seems to be around 0.25-0.3. On the other hand, depending on what you're doing - restyling, adding elements, changing expressions - that wild creativity isn't always a detriment. These models are also fairly decent for inpainting with no additional tools, so if there's just one or two parts of the picture you don't agree with then slap the thing into inpaint, set the mask blur to something like 20 (to give the inpaint area a gradual border), [Whole picture], denoise to 0.35-0.4, Resize by to x1 (to keep the overall picture at the same resolution) - and re-roll a few times.

Once you've got your final result, but some areas are still lacking, switch that up to [Masked only], set the resolution to 1024x1024 (SDXL's default) and modify the prompt to contain only the elements inside your inpaint region. This'll take the region, scale it up to 1024x1024, regenerate at that resolution, scale it back down and re-insert it into the original picture, resulting in "sharper," more detailed elements in that area. It's how I do faces - and how ADetailer does them automatically, if you're willing to spare the compute/fiddle with its settings. Also handy for hands, small accessories that don't quite resolve though other elements are fine, and so on.
>>
>>48204273
https://files.catbox.moe/sf0twk.png
https://files.catbox.moe/e97r5j.png
https://files.catbox.moe/65uzi4.png
https://files.catbox.moe/dsxn4d.png
>>
>>48204499
Why am I black?
>>
>>48204510
We're 100% black.
>>
File: 6003571546.jpg (977 KB, 1947x1404)
977 KB
977 KB JPG
>>48204510
>>48204514
>>
>>48204496
>oppai yatsuhashi
I'm interested, tell me more
>>
>>48204510
She's feeding the entire world, not just you.
>>
>>48204514
Me too.
>>
>>48204497
How do you prompt the glasses falling like that?
>>
>>48204496
>>48203129
Good zankurohus thank god for ai since I think the original artist just drew like suika once and that's and his style is so nice
>>
>>48204496
>>48206524
lmao the real Zankuro already bent the knee to AI and became a slopper himself.
>>
>>48206549
Did he really?
>>
>>
>>
>>48206594
Is she ok
>>
>>48206549
What he uses aisloppa for doe? Inspiration or just tracing over it?
>>
Why does every Zankurohu look like they were forcefed nothing but mashed potatoes and gravy for a year straight
>>
>>48206572
This one is cool
>>
>>48204727
Same way you prompt a background manga character with nowhere near enough material to train on.
You don't; it's a manual edit.
>>48206610
She just spilled some ink. Not an uncommon occurrence in her line of work.
>>
File: nu.png (980 KB, 1024x1536)
980 KB
980 KB PNG
>>
>>48206719
Have you never looked at a Zankuro drawing in your life?
>>
>>48206549
If you can't beat 'em, join 'em.
>>
>>48206549
Well, why not? If you've got a Fanbox quota to fill, AI is an invaluable time saver.
>>
File: 00004-454415361.png (787 KB, 800x1296)
787 KB
787 KB PNG
>>48204514
Yes, all of us. Undoubtedly. We all are.
>>
>>48208747
*pours acetone on you*
>>
File: boatimage.jpg (402 KB, 2042x1366)
402 KB
402 KB JPG
>>
>>48206565
>>48206628
Not sure what the other anon is referring to but I know he was curious about using AI as a tool when he saw someone trained a LoRA on his style way back when around the first NAI leak. Haven't heard of him using it recently and he sometimes puts that godawful glaze shit on his pics so ???
>>
>>48210241
He still condemns it, I don't know what that other anon is saying. https://x.com/zankurokxiiai
>>
>>48210239
Is this flux?
>>
File: pache.png (2.07 MB, 1472x1472)
2.07 MB
2.07 MB PNG
>>48206719
they are squishy
>>
>>48211314
obesity
>>
File: 00031-1268635907.png (1.09 MB, 896x1296)
1.09 MB
1.09 MB PNG
I for one see 0 harm in using an artist's artstyle to draw something that he was never going to end up drawing anyways. If anything, they should be flattered that I like their artstyle so much that I want to see it applied for other things outside of what they draw.
>>
>>48211401
Like taking an NTR or bestiality artist and using their style to draw vanilla loving sex for the sake of procreation
>>
>>48211445
Same in reverse.
>>
>>48210307
Huh, I didn't know about this account. His Tweets seem oddly conflicted, like he's still getting upset about it even though he's generally acknowledged he can't do anything about it. And he's clearly continued toying with it himself, plus I do remember him doing some gens way back and redrawing them...
>>
>>48210239
Damn nice stuff, but those trees just to the top and left of Seiga desperately need fixing.
>>
>>48210239
Model?
>>
>>48211314
More fatchouli please.
>>
File: SYXxCi.png (342 KB, 768x1024)
342 KB
342 KB PNG
Hmm.
To work on my Yuuka, or try making a new bot entirely...
Thinking a gymbunny, a Reisen based off pic related
Reisen with the classic /fit/ lesson baked in, that no matter how muscular and shredded she gets, she will always be a spineless, cowardly doormat
>>
File: IMG_8285.jpg (303 KB, 1024x1536)
303 KB
303 KB JPG
>>48212072
You can tell I got tired of trying to fix the details with an inpaint and just used a filler. Same goes for hands. Inpainting takes too much time and results are not worth it.
>>48212418
Stablediffusion
>>
Can someone gen 2hus fellatio in hero neisan style?
>>
>>48204244
Thank you! I love them
>>
hungry, vicious chupa
https://files.catbox.moe/roxwye.jpg
probably no longer hungry fish
https://files.catbox.moe/w6qqys.jpg

>>48211401
Artists are touchy enough without perceived plagiarism in the equation. And when you add the various fantiaboxtreons into the mix, it's no longer plagiarism - but active, monetary competition. In their eyes, anyway. Can't blame them anyway. The Japanese are notoriously tech-illiterate, and it's not like the field isn't rife with misinformation by itself.
>>
File: byakuren.png (1.94 MB, 1088x1920)
1.94 MB
1.94 MB PNG
>>
>>48218767
She looks dangerous
>>
>>48219285
Of course she does. It's Byakuren.
>>
>>48218767
M-mama...
>>
>>48174743
assaults little village shotas
>>48188907
assaults ojii-sans
>>48218767
assaults healthy young males
>>
Moar wriggle onegai.
>>
File: model_thing.png (40 KB, 1679x689)
40 KB
40 KB PNG
I just wanted to ask about the addon that shows prompts embedded in a picture not showing the name of the model, just wanted to know if I'm blind or missed something.
>>
>>48220421
I guess I didn't implement it to show the model name, but because that's NAI metadata there's only two likely possibilities it could be (anime or furry) and with the image for context you can deduce it.
>>
File: 1722123268209938.png (2.35 MB, 2000x4000)
2.35 MB
2.35 MB PNG
Hoping for some adult Suwako and probably in a romantic scenario with Rinnosuke.
>>
>>48220594
>Rinnosuke
...Why?
>>
File: redraw7.png (1.47 MB, 1352x1736)
1.47 MB
1.47 MB PNG
https://files.catbox.moe/t5q1oj.png

a serious looking auntie (probably in her prime-time days), who's not happy at being disturbed...

still fresh at AI genning. Second real attempt at a workflow. Trying to shorten things down without getting too hung up on small details like her tile.
>>
File: 54756865485.png (34 KB, 1112x118)
34 KB
34 KB PNG
>I'm not shy anymore
>Rika giggles nervously
https://files.catbox.moe/6b7k0n.png
>>
>>48220712
He's a cuck, a Rincuck. Some people in the fandom want to get cucked by the few male characters in the setting.
>>
>>48220960
Here she is without clothes

https://litter.catbox.moe/7dfvq1.png
>>
>>48220594
Same energy.
>>
>>48221340
How do you gen the exact same pic without clothes?
>>
>>48221340
Her breasts grew in size!

Wasn't asked for, but appreciated all the same, anon. (I imagine someone would've asked for a nude inpaint sooner regardless.)
>>
It's just inpainting the original image and adding in shit that I didn't want to see in the negative prompt and stuff I did want to see in the positive prompt. In this case my horny porn addled brain just put in (large breasts, nipples, pussy) in the positive prompt and (clothes, dress, skirt) in the negative one.
>>
File: e543543543543.png (62 KB, 204x253)
62 KB
62 KB PNG
>>48221350
Nigga i get the Suwako image is weird, but the tumblr abomination you posted is pure distilled warp energy. Twisted demonic putrid eldritch energy that only Lovecraft could dream up. It's probably what half the people in this board wishes they would look like.
>>
File: Namori Arisu.png (1.32 MB, 832x1216)
1.32 MB
1.32 MB PNG
>>48211445
You lack imagination, anon.
>>
>>48221622
My power level was hidden. Nobody is equipped to handle my imagination at full power, not even myself.
>>
Any tips on making characters having sex with each other? I tried using ControlNet but it didn't work great for most positions.
I generated a lot of Yakumo futa, and it's a waste for their dicks to go unused.
>>
>>48221683
have you considered prompting for it
this isn't sd1.5 anymore
>>
>>48221729
NTA but there's no model that doesn't sucks at prompting more than 2 characters interacting, even the mighty NAI barely can.
>>
>>48221352
Img2img trace her entire body from the neck down, then use tags like nsfw, nude, nude filter, nipples, etc.
>>
>>48221100
I like it when chatbots are accidentally funny like that
>>
>>48221622
Why does it look so blurry?
>>
Do Noob devs filter out the artists who don't consent to their art being used to train AI? Morinohon has recently started putting this watermark on his art.
>>
>>48225312
Hopefully not, I doubt it
>>
>>48222575
But the dude is asking for sex which is usually 2 just 2 entities. Both NAI and noob can handle that for sure unless the characters are very obscure.
>>48225312
No, they don't care, nobody except for the pony creator does.
>>
>>48225312
Where does this mindset of "if I tell them not to, they won't" come from?
>>
>>48226810
The japanese unironically think the rest of the world is as considerate as they are.
>>
File: declutter.png (2.88 MB, 1731x1146)
2.88 MB
2.88 MB PNG
>>48225312
>>48226810
>>48226909
Artists are so unaware
>>
>>48227430
@PIXIV
@FBI
@POTUS
@FANTIA
DO SOMETHING!!
>>
>>48227430
>FOUR watermarks and a gigantic cryfest on the first
good lord
>>
File: hecc.png (2.41 MB, 1088x1920)
2.41 MB
2.41 MB PNG
>>
>>48227430
deserved, fuck these entitled little shits
>>
File: AI lineart extraction.png (1.96 MB, 2233x633)
1.96 MB
1.96 MB PNG
>>48227430
I forgot about that function. It's neat stuff.
>>
>>48227430
>>48227920
lmao artists are fucked
>>
>>48228003
Is it good thing though?
>>
>>48227920
How does regular edge detection compare?
>>
>>48227430
>>48227920
What are you guys using, is that NAI or local?
>>
>>48228505
NAI using their "director tools"
>>
>>48220594
Bro what is this monstrosity she looks like a JoJo character
>>
>>48227430
What about the low transparency ones?
>>
File: 1618116607684.jpg (749 KB, 1200x984)
749 KB
749 KB JPG
>>48229755
>>48220594
She looks more like a Captain Tsubasa character.
>>
File: 6232619689.jpg (473 KB, 1254x1771)
473 KB
473 KB JPG
>>48225312
>>48227430
Interesting.
>>
What are some of the most memorable moments you guys have had with 2hu chatbots?
>>
>>48172121
Why are you still spamming Koishi threads in /g/?
>>
what loras are compatible with noob?
>>
File: 464367347.png (278 KB, 408x377)
278 KB
278 KB PNG
the 10gu
https://files.catbox.moe/k55eag.png
>>
inchling here
>>
>>48232780
not mine but
https://desuarchive.org/g/thread/101837772/#101840369
unhinged sizemaniac has his fun
>>
>>48235840
Where? *looks around, searching for the source of the voice*
>>
A cushy job posting at the SDM where all you do is simple accounting. Sometimes the younger mistress comes by to bother you, but all you need to do is wag your fingers a little to make her go away:
https://files.catbox.moe/renle5.jpg
Sekiwaitress:
https://files.catbox.moe/myazfp.jpg
Futo showing her best side:
https://files.catbox.moe/6a767r.jpg
Koakuma's read-aloud session:
https://files.catbox.moe/okvj0s.jpg

>>48235679
SDXL LORAs have a chance to work, but it's a crapshoot. IL-based LORAs should work out of the box. There's a filter on Civit.
>>48232780
Nothing out of the ordinary, but I once had Sekibanki veer into a story about how she used to know Reimu when she'd been younger and part-timed at Geidontei. With the usual twist for lewdslop models. Built an entire backstory on how Reimu would act at work, seduce the customers, etc. etc.
... The kicker is that my avatar in that chat was Reimu's son.
>>
>>48232780
Oh that's easy! I roleplayed an isekai scenario back when Claude 3 had just come out: I put myself in the bamboo forest at dusk. Instead of receiving help from random 2hus (I did with earlier models), Claude made Tewi appear out of nowhere and start playing with me like I were her prey. Some of the insults stung so bad I felt them through the screen and to call it a day.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.