[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] [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
/aco/ - Adult Cartoons


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1731215730791923.jpg (107 KB, 1024x1024)
107 KB
107 KB JPG
Previously, on /asdg/ >>8944432
The cartoony side: >>>/aco/csdg

No DALL-E, Meta AI or hailuo AI generated stuff.
No futa.
>>>/aco/futasdg
Report trolls.

>Guides
More modern options:
https://github.com/comfyanonymous/ComfyUI
https://github.com/lllyasviel/stable-diffusion-webui-forge
Too old, might hold some value: github.com/AUTOMATIC1111/stable-diffusion-webui/wiki | rentry.org/sdg-link | rentry.org/rentrysd
Hires Fix: rentry.org/hiresfixjan23
Inpainting/Outpainting: rentry.org/drfar | rentry.org/inpainting-guide-SD
Animation: rentry.org/AnimAnon
Create LoRAs, TIs: rentry.org/59xed3 | rentry.org/simplified-embed-training
>Auto1111 extensions
ControlNet: github.com/Mikubill/sd-webui-controlnet | rentry.org/dummycontrolnet
Regional Prompter: github.com/hako-mikan/sd-webui-regional-prompter
Tag autocomplete: github.com/DominikDoom/a1111-sd-webui-tagcomplete

4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
4chan-x userscript from /h/ that lets you read generation metadata from images in-browser, and upload to catbox with one click when you want to share metadata.

>Wan Video 2.1
Setup Guide: https://rentry.org/wan21kjguide
Triton Guide: https://github.com/woct0rdho/triton-windows/blob/readme/README.md
Easy Sage Attention Install: https://github.com/woct0rdho/SageAttention/releases

>Related
>>>/b/degen
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/g/sdg
>>>/trash/sdg
>>>/trash/slop
>>>/vg/aids
>>>/vt/vtai
>>
File: bwft7z4xlybd1.jpg (617 KB, 2544x3720)
617 KB
617 KB JPG
>>
File: 5otvpfgtukqd1.png (1.6 MB, 1664x2432)
1.6 MB
1.6 MB PNG
>>
File: 9mv3wnwfx3od1.jpg (112 KB, 920x1536)
112 KB
112 KB JPG
>>
File: 1726265850747298.jpg (139 KB, 1440x1120)
139 KB
139 KB JPG
>>
File: 1729751959739626.png (793 KB, 720x1080)
793 KB
793 KB PNG
>>
File: 1723026382219716.png (2 MB, 1248x1608)
2 MB
2 MB PNG
>>
File: 0.png (243 KB, 392x391)
243 KB
243 KB PNG
I don't remember seeing these images posted in the last couple of threads.
Are you... reposting AI sl0p from other websites?

No one can can be that deranged right. Are you picking ugly images on purpose. Why wouldn't you repost something like >>8963072
This seems too stupid to be a troll, and yet.

*i just had to delete two typo'd posts from the mental flinch of finding out you're reposting r*ddit patreon tier AI slop.
>>
>>8964431
Weird example to pick, but otherwise I agree.
>>
>>8964431
You will never win against spergs and schizos. Look at /h/ and /g/.
>>
>>8964455
How is it weird? Are you an AI slop critic? It's more high-effort than other things posted in that thread. Also, are you the OP agreeing with yourself about being a lunatic?

How many months have you been making this thread? Is it years?
>>
>>8964461
Why are you angry even after I said you're right? Calm down bro, you'll get an aneurysm.

I mean it's a nice gen but she looks like a plastic doll. Maybe something to post a few images further down, not to represent the photorealism thread.
>>
>>8964481
props to you my nigga. never back down, always deny, never apologize.
>>
File: 00028-3933726104.jpg (166 KB, 1824x1248)
166 KB
166 KB JPG
>>
>>8964481
You didn't even place the word "photorealism" in the OP you fucking farm animal.

Spergs like you is how boards end up with 4 AI generals.
>>
>>8962615
Yeah that model got a bias for rounder breasts there. This should be better, but I can try again in a few days
>>
File: comf_00070_.png (3.23 MB, 1280x1920)
3.23 MB
3.23 MB PNG
>>
>>8964613
God bless you, anon. Githbox?
>>
Requesting the RWBY girls and Velocity from Cyberforce turned realistic please.

Including Pyrrah, that redhaired amazonian warrior.
>>
File: 20250830024755-2327282484.jpg (642 KB, 1664x2432)
642 KB
642 KB JPG
>>8964714
>Velocity from Cyberforce
I can do just her because I happen to have a LoRA for her.
>>
File: 985712942.png (2.32 MB, 1440x1920)
2.32 MB
2.32 MB PNG
>>8964431
The fact that he does it to two generals and pretends he's doing nothing wrong is funny.

But you should have seen the meltdowns from /aco/ posters when jannies weren't banning AI threads in 2022-3. This whole website is so far gone from that moment.
>>
>>
>>8964469
>you're right OP's a fag, but you could've picked a better example image to show it
>ah-ha! clearly you must be the OP
>>
>>8964431
Probably because that picture belongs in csdg? now post wet ai tits or fuck off.
>>
File: ComfyUI_temp_ocgub_00009_.png (1.45 MB, 1024x1024)
1.45 MB
1.45 MB PNG
>>8964714
It's Pyrrha. Probably to foreshadow their pyrrhic victory at the end of season 3.
>>
>>8964916
Your video is 2.5D like that image, except hideous and retarded looking.
Are you pretending or is it brain worms.
>>
>>8964916
It's borderline. I think it's at least trying enough to be here.
>>
>>8964741
That big smile with slightly mis-sized teeth is perfect!

A minor quibble would be that her hair would be swept back from the wind effect but other than that I love it!
Any more?


>>8964879
Are thse from a video game?

>>8964921
Excellent!
(wait why am I chanelling Bill and Ted? Eh, oh well)
Can hardly wait to see the rest of the cast!

Oddly enough I even want to see what kind of situation a drunk half naked Krow gets himself into.
:)

After you do the indiviguals could you do a group shot of the red-head brigade [Penny, Pyrrah, Nora, and yes even Ruby herself{{even if it is only the tips that she dies it still works for her}}]?
>>
>>8964741
>>8965030
>Any more?

Don't get greedy , you do not want to annoy the genner.

>>8964921
>>8965030
>Can hardly wait to see the rest of the cast!
Again do not get greedy or annoy the genner.

>>8965030
>Oddly enough I even want to see what kind of situation a drunk half naked Krow gets himself into.
>:)
.....okay so I just said not to get greedy but now I want to see a human Krow naked in a birdbath.
U R a bad influence on me.
...and some how I am seeing Ren and Juan spending quality male bonding time in the men's side of an onsen. No not having sex just hiding out from the girls!
If any genners try to actually creat this scene you get bonus points if the girls are peeking overr the fence.


>>8965030
>After you do the indiviguals could you do a group shot of the red-head brigade [Penny, Pyrrah, Nora, and yes even Ruby herself

RWBY does havea fair amount of redheasds dont it?
>>
>>8965037
>you get bonus points if the girls are peeking overr the fence.
Naw, Yang would lead them on a reverse panty raid!
:)
:)
:)
>>
>>8964741
>Velocity from Cyberforce
>I can do just her because I happen to have a LoRA for her.

You just HAPPENED to have one huh?
yeah right, just admit that you have fifty gigs of Velocity gens with you gueststarring as her boyfriend in them.

Oh, and whileI do like this a lot , her boobs are just a little to big, remember that this is Velocity not a fan-ballooned Powergirl.
>>
>>8965037
I am visualizing Goodwitch in the mixed gender baths with Professor Port and the one with the wild hair.
>>
>>8964879
Who are these characters?
Or are they just random gens?
>>
>>8965058
Can never have enough tits
>>
>>8964921
If the shot was just bigenough then I am sure that we would see that Jaunnes face has a big silly grin on it!
>>
>>8964921
Pyrrah nude versus Pyrrah in leather romanesque armour?
I, I cannot deside which I like more!

More please, of both , I need a tie breaker!
>>
File: instagram-serog9876.jpg (154 KB, 672x864)
154 KB
154 KB JPG
>>8964283
>>8964507
>>8964613
>>8964921

Nice
>>
>>
>>8965144
Easy to tell you're samefagging when you keep misspelling her name.
>>
>>8965030
>>8965060
The characters in the cyberpunk image are randomly generated. I was trying out Sam Base, but I'm not really satisfied with it so I'm going back to PornMaster-Pro.

>>8965058
I HAPPENED to have a Velocity LoRA because I don't actually care that much about her, I'm really interested in reproducing Kenneth Rocafort's artstyle from her series, but that's probably off-topic for this thread.
>>
>>8965030
>After you do the indiviguals
Might be my weird realism workflow but it takes more effort than just writing a prompt. I'll definitely get to them at some point, I would even if you hadn't asked. Already done a bunch of RWBYs before; just in that awkward uncanny pseudo-realism cause back then I couldn't do better.

They are kinda funny to work with though. Not only do you get the usual anime cosplay issue where the face doesn't quite fit, but they're also too young in canon so you have to make them look older. And then Ruby has white eyes, Penny's a robot...is she supposed to look fake? I guess in-universe people couldn't tell by looking at her.
>>
>>8965168
If you end up doing more, could you show her "feet" as well?
>>
>>8965166
thanks, but what's with the instagram filename
this is like a drawthread for AI,not an imagedump thread. post your own stuff
>>
>>8965203
Guess it depends on what you're doing. For me Pornmaster too quickly reverts into plasticky 3D if I prompt for a named girl or use a lora.

By they way, look into Adetailer or inpainting to fix their eyes. It's a small thing but makes a big difference.
>>
>>8965210
I did use Adetailer. The problem is, I put too much weight on "single mechanical eye" because that usually works on anime artstyles.
>>
Ah my bad, I thought it didn't look intentional.
>>
>>8965209
that's my images and my insta
>>
>>
feast your eyes
>>
File: Unknown-10.jpg (261 KB, 896x1152)
261 KB
261 KB JPG
>>8965208
Hooves or real feet
>>
>>8965274
She's draenei, right? Hooves of course.
>>
How can you generate NSFW stories?
>>
>>8965285
Run an LLM locally?
>>>/g/lmg/
>>
>>8964699
Easier to share the metadata invoke shows.
https://hastebin.ianhon.com/ebea

Shit's great for inpainting but it doesn't save meta as plain text, it seems to embed it in some channel only Invoke knows to read from
>>
>>
>>8965450
i think i saw this pic (the drawn version) today lol
>>
File: 20250825084554-3277693356.jpg (677 KB, 2432x1664)
677 KB
677 KB JPG
>>
>>
File: 0_.webm (1.04 MB, 592x792)
1.04 MB
1.04 MB WEBM
>>
>>8966160
>>>/aco/futasdg
>>
>>
>>
>>
>>8965220
holy crap, I can feel that coarse hair rubbing against my mouth such detail
>>
>>8966282
How about you feel the coarse hair on my balls
>>
>>8966191
If being an illiterate retard was a crime, you'd have been long disappeared.
>>
>>
>>8966301
fuck off

I don't know what happened to the containment thread but it still says no futa in the OP, take your tranny shit somewhere else
>>
>>8966317
>tranny
I hope you're not the one posting gens with lumberjack chins and anvil jaws, because that would be ironic for you.
>>
>>
>>8966332
I don't actually know if I am, by that description. Could you point to one of those gens?
>>
>>8966334
It's pretty obvious if you know what women are supposed to look like.
>>
>>
>>8966302
>>8966276
>>8964613
>>8964794
any chance of a catbox on any of these? great gens!
>>
File: download (22).png (1.83 MB, 896x1152)
1.83 MB
1.83 MB PNG
>>
File: download (17).png (2.06 MB, 1631x1088)
2.06 MB
2.06 MB PNG
>>
File: download (9).png (1.98 MB, 832x1248)
1.98 MB
1.98 MB PNG
>>
File: download (12).png (2.11 MB, 736x1408)
2.11 MB
2.11 MB PNG
>>
>>8966160
Fuck off with that shit faggot.
>>
File: 20250901023638-1716002858.jpg (1.04 MB, 1664x2432)
1.04 MB
1.04 MB JPG
>>
File: 00200-1863805872.jpg (106 KB, 1824x1248)
106 KB
106 KB JPG
>>
File: 00065-2992594256.jpg (157 KB, 1536x1536)
157 KB
157 KB JPG
>>
>>8966871
who's the girl?

>>8966872
I always thought the suits would be shiny latex
cool nep though, first attempt I saw at doing the wings too
>>
>>8964269
Where do you guys find good prompts for images? Civitai used to be good but now every top images is just some gay shit bots upvote to be on the front page
>>
>>8967329
The front page is actually people paying to get their images featured. There's an ongoing auction for feature slots, anyone can participate.

Have you tried reading the prompts you've stolen up till now? Figure out what makes them good in the first place. Then start experimenting with your own, that's what most people do.
>>
>>8967329
Front page was always garbage. You´ll have to find some collections like https://civitai.com/collections/7175002 And as >>8967333
says, reverse-engineering prompts and experimentation is step 2.
>>
>>
>>8966258
>>8966263
nice!
>>
>>
>>8967644
God yes
More Gilf/Mature
>>
>>8964269
Anyone had any luck with this?
https://myaiforce.com/comfyui-ipdapter-swap-outfit/

If so, I have a follow up question, how flexible is it with that it considers clothing?

Like if I submit the nude body of a big booba thiccie pornstar or something as the "clothes" will it give the subject that big booba thiccie body? Can it basically be used like Deepfake but for bodies?

Or does part of its training specifically make it try to ignore body parts as much as it can so it can only focus on the clothes?
>>
>>8966368
Here's the original with meta but only Invoke seems to be able to read it. Shared the meta here >>8965444
>>
>>8967790
Brainfart, here cat https://files.catbox.moe/i2cyj5.png
Also for this one https://files.catbox.moe/y544ly.png
>>
File: test.jpg (1.81 MB, 2450x3700)
1.81 MB
1.81 MB JPG
>>8967702
There's two parts to this, both made to sound much better than they are in practice.

First part is the automatic masking. Cool idea, but you can draw the mask yourself in fifteen seconds and not need a 20-page tutorial. When you look at what was achieved, it only changed one dress for another of a different color but very similar cut and the same shape. If you wanted to give her jeans and a shirt instead, or make her nude, it will sometimes fail to blend with the rest of the image. But yes, we can already do this with regular inpainting, picrel.

The other part is IPAdapter, using a reference image instead of prompting for what you want. In theory IPadapter can replicate anything that a zen prompt master could prompt for, if he knew all the ins-and-outs and little quirks of the checkpoint by heart. Information from the reference image is encoded the same way a text prompt would be. When you look at the dress in her result it's not the same color or pattern, they're just similar. Now, the problem is you're also getting the shading, her specific skin tone, some pose influence, her hair...there's no way to separate just one concept, and all of that excess information makes the blending issue I mentioned much worse. It's better suited for generating whole images than for inpainting. Compare prompts in the top row to references in the bottom row.
>>
File: 20250901054122-654718941.jpg (533 KB, 1664x2432)
533 KB
533 KB JPG
>>
File: 00432-2217932696.jpg (192 KB, 1440x1920)
192 KB
192 KB JPG
>>8964269
made a femshep set (wan animated):
https://civitai.com/posts/21701487

there's another femshep set there too, working on making more, eventually Liara too
>>
>>8967794
hmm. but if you wanted to full automate inpaint nudifying or breast enlargement or whatever else, it would probably be the best option, since you wouldn't need a human contributor to draw the masks.

So sloppa. It has a use case, but the use is sloppa. Mass produced inferior goods. As per usual, human involvement is required for quality.
>>
>>8968101
Apparently the femShep model was completely human-created and so it has a slightly surreal quality. I'd wondered how a realistic interpretation might look. This is convincing.
>>
>>8968153
>Apparently the femShep model was completely human-created and so it has a slightly surreal quality.
Is this a bot? You type like a fag. A fagbot.
>>
>>8968179
more likely ESL
>>
>>8968179
People who have fathers in their lives tend to be a bit more eloquent.
>>
>>8967681
>>
>>8964282
How did you make this? I downloaded some stuff in the Original Post and got nothing like this.
There must be a work flow, where you make a crude start and then fix eyes, arms, etc.
Is this from a pay for service website or do you have a decent NVIDIA card?
>>
>>8968512
it's an iterative learning process. you start bad then get good. post some of the parameters or models you're using and you could get advice. if you're feeling brave post a pic/prompt
>>
>>8968527
I have a NVIDIA GeForce RTX 3070 Ti (Laptop)
Is that enough? or do I need to go to a website?
>>
File: 00067-2324667211.png (326 KB, 512x512)
326 KB
326 KB PNG
This is about the best I was able to do. I don't recall the model or the prompts, and I used a Lora for Taylor Swift....
What did I do wrong?
>>
>>8968550
Well, it would be helpful if you recalled the model. At least whether it was based on SD1.5 or SD-XL.

In the case of 1.5, you need to upscale and run another AI pass over it. They call the process hires-fix. In the case of SDXL, you should be using resolutions of 1 megapixel (1024x1024) for the base image.
>>
>>8968550
That looks like done with SD 1.5 at base resolution, nothing but the LoRA.
For starters, use a better checkpoint like BigLust. Then use a bigger resolution and hires.fix or equivalent.
>>
>>8968558

From PNG Info...
===
parameters
taylor swift, naked
Negative prompt: extra fingers, extra arms
Steps: 30, Sampler: Euler a, Schedule type: Automatic, CFG scale: 7, Seed: 2324667211, Size: 512x512, Model hash: 2ab7fa3211, Model: real-dream-8, Version: v1.10.1
===
If I go larger than 512x512, it crashes. (I think, I'm giving it another try). I thought that things were made in 512 size, then upsized...
>>
>>8968560
Well it's not like 1.5 is bad for just 1girl standing. But it needs to gen at 512, which means upscaling is absolutely required.
>>
>>8968563
>I thought that things were made in 512 size, then upsized...
On that model, yes. And they look awful before it's done, it's just like a rough draft for the AI to follow while upscaling.
>>
File: 00007-4245557328.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
parameters

==taylor swift

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 4245557328, Size: 1024x1024, Model hash: 07b985d12f, Model: SDXLFaetastic_v24, Version: v1.10.1
==
okay, I was wrong. it didn't crash.
I guess I just need to keep trying. You guys are so good at this.
>>
File: 1686551261251046.jpg (750 KB, 2466x1804)
750 KB
750 KB JPG
>>8968564
Then he should play with the image resolution, getting a more useful ratio than 1:1 to get the most body possible into the image. And go upper body/portrait to get a more detailed face.
>>
>>8968568
Thanks for the resource it will be useful
>>
File: The Shoes 3 1.png (1.99 MB, 1512x1368)
1.99 MB
1.99 MB PNG
>>
>>8968495
>>8968600
Hell yeah
>>
File: The Shoes 1 2.png (680 KB, 664x896)
680 KB
680 KB PNG
>>8968777
>>
>>8967545
>>8968101
This place is kinda... JuicY!
>>
>>8964269
adult noodle is so busted
>>
File: ComfyUI_14510_.png (2.84 MB, 1248x1824)
2.84 MB
2.84 MB PNG
>>
>>8968874
trying to cover those great big bazongas is a losing battle
>>
>>8964269
What are the best realism models locally? Prompt adherence a plus, but if it can at least be wrangled, good. I need to generate teenage PAWGs getting fucked in the ass so I can turn them into videos. Each model I've tried seems to have something I like, but not all. I need a master key of PAWG generation.
>>
>>8968568
>no 832x1216 or 1024x1024
:(
>>
>>8968910
Without knowing what you’re doing wrong we can’t really prescribe anything. List some models you’ve used and some prompts you’ve tried
>>
>>8968919
It's not really strict like that from what I've seen, as long as you stay near 1 megapixel for the base gen you'll be okay regardless of aspect ratio. Going higher increases the chance of extra fingers and other broken anatomy, the higher you go the worse it gets. Going lower is a waste of detail.
>>
>>8968910
There's probably not one "best" one, to me it seems like all the nsfw tunes use limited and biased datasets so the model ends up being good at one things and worse at others. Almost like a bloated 7gb Lora. I primarily use SamBase because all I do are anime girl cosplays, someone else might use Lustify or Biglust because they focus on hardcore that looks like mainstream porn scenes, someone else might want amateur stuff, or insta slut pinups, or bimbo aesthetic, clothes and sfw poses, etc.
>>
File: sdxl-resolutions.png (266 KB, 2048x1320)
266 KB
266 KB PNG
>>8968919
>>8968941
I usually refer to this chart from the paper announcing the SDXL base model, which lists the resolutions that SDXL was trained at.
>>
>>8968922
>>8968941
>>8969027

Lustify (XXL) seems to have no idea what anal even is. The asses are amazing though, but I can't get any anal.

RealismPony by Yogi or whatever it's called has good lighting and girls, but mutant dicks and bad sex positions.

All the Illustrious models I've tried for realism either do realism bad, or are just too generic and sanitized looking, for lack of a better word. Which is weird because Illustrious is perfect for my animated stuff, but realism I haven't found a good checkpoint for.

Those are the ones I've tried. Obviously Flux would be awesome if it did NSFW well, but even the NSFW loras and checkpoints I've tried seem to be worse than just Pony.
>>
>>8969280
Have you tried Sam Base?
>>
>>8969284
At work now but I'll look into it and check back.
>>
>>8969280
That's exactly what I meant. You may end up with a half dozen checkpoints each good for different things.

Check for metadata if you see a catbox itt, or in the previous ones. Easier to find decent checkpoints that way than trying stuff off Civitai at random.
>>
I've played around with realistic checkpoints and Cyber illustrious is impressive but kinda same facey and reluctant to face prompt. Any good checkpoint with face variety? Or tips for faces beside using Loras.
>>
>>8969425
If you're using adetailer or you're inpainting faces yourself, you can switch checkpoints just for this part. Get access to another checkpoints' face variety.
>>
>>
>>
>>8969436
I though about adetailer, I'll try thanks. Any idea for a good checkpoint for this? I probably should look into SDXL ones for this right?
>>
>>8969490
Guess so, or Flux. They don't even need to be nsfw if it's just for the face.
>>
>>
>>
>>8969512
gadzooks
>>
File: 00007.png (3.86 MB, 1664x2432)
3.86 MB
3.86 MB PNG
>>
>>
>>
>>8969528
Ay Carumba!
>>
>>
>>
>>
File: 00016-4110021380.png (614 KB, 512x1024)
614 KB
614 KB PNG
>>8969280
>Lustify (XXL) seems to have no idea what anal even is.
Great. That's a feature.
But 50,000 buzz.... yeah.. $50. or wait two weeks. I'm not there yet and need to learn more about inpaint and other stuff.
Maybe when I get past my beginner's phase..
>>
File: ComfyUI_00055_.png (1.45 MB, 896x1184)
1.45 MB
1.45 MB PNG
>>
File: ComfyUI-RIFE_00518_.webm (3.78 MB, 720x960)
3.78 MB
3.78 MB WEBM
>>
File: Image 15.jpg (206 KB, 768x1496)
206 KB
206 KB JPG
>>
File: IMG_2148.png (1.12 MB, 768x1152)
1.12 MB
1.12 MB PNG
>>
File: IMG_2179.png (1.48 MB, 1024x1024)
1.48 MB
1.48 MB PNG
>>
File: IMG_0417.png (1.29 MB, 568x1744)
1.29 MB
1.29 MB PNG
>>
>>
File: elf hugger_00540_.png (953 KB, 600x1200)
953 KB
953 KB PNG
>>
So does it HAVE to be exclusively realistic or 3D-ish or is it just what's taken over the General these days? I swear I read weeks ago it wasn't just for realism.
>>
Either way, have two.
>>
File: 00084-2537493227.png (2.89 MB, 1248x1824)
2.89 MB
2.89 MB PNG
>>
File: 00003-2775434457.jpg (107 KB, 1824x1248)
107 KB
107 KB JPG
>>
File: 00128-4047045188.jpg (155 KB, 1824x1248)
155 KB
155 KB JPG
>>
File: file.png (521 KB, 768x512)
521 KB
521 KB PNG
anyone knows why SD have these weird eyes every time there is more than 2 characters in the picture and how to fix it? i assume its SD, from perchance
>>
>>8971594
In my limited experience with perchance I've found adding detailed eyes to the prompt and bad eyes in the negatives bar helps a little but it might just be placebo
you might try asking in the mmai thread but it seems like the celeb posters have taken over and they don't use perchance
>>
File: file.png (519 KB, 768x512)
519 KB
519 KB PNG
>>8971613
i dont think its working when its a complex scene (or complex for this shitty generator anyway).
>>
File: file.png (639 KB, 768x512)
639 KB
639 KB PNG
>>8971629
it can have them 90% fine when its just characters standing looking forward
but otherwise i would have to reroll and hope that the cursed ones have their eyes closed or face turned
>>
>>8971629
hands im already used to becoming this horrific mess. Gemini can fix those alright most of the time but it will refuse if its porn or if it can guess that its porn.
>>
>>8971629
Yeah, I've never gone for a scene that ambitious so I wasn't sure. Sorry it didn't help
>>
>>8971484
love your stuff
is the blonde with the condom in her mouth earlier in the thread yours too?
>>
>>8971458
The cartoon thread next door accepts everything else. Unless it's strictly anime, in which case there's /h/.
>>
>>8971594
More people in the scene means they're all further away from the camera. >>8949413

You're also using a weird tiny resolution, SDXL runs best at 1 megapixel, before upscaling. If you are using old SD1.5, doing an upscale and second AI pass is a must. Tutorials usually call the process HiresFix.
>>
File: IMG_20250905_114954.jpg (392 KB, 832x1216)
392 KB
392 KB JPG
>>
>>8972061
i have no control over such things, perchance is a web interface
>>
>>8971349
This is my favorite thing Ive ever seen
>>
File: 00008-2650166374.png (347 KB, 512x512)
347 KB
347 KB PNG
Let's say you're broke.
Let's say you hate YouTube ads.
What's the best way to learn Stable diffusion, without shelling out a lot of money
>>
>>8972353
maybe learn how to remove ads first
>>
>>8972108
model?
>>
>>8972353
Civitai.com shows prompts and settings on almost all images on the site. Anytime you see a catbox link in an AI thread, it likely has generation metadata. Load up an existing image and start messing with it, figure out why it works and what you're doing differently.

Then once you have some experience you'll know how to ask better questions.
>>
File: ComfyUI-RIFE_00573_.webm (2.78 MB, 720x960)
2.78 MB
2.78 MB WEBM
>>
File: 00020-2045721406.jpg (163 KB, 1248x1824)
163 KB
163 KB JPG
>>8971795
yes those were mines
>>
File: 00003.png (2.23 MB, 1248x1824)
2.23 MB
2.23 MB PNG
>>
>>8972829
very cool
that blonde definitely helped me out the other day
good stuff
>>
>>8972392
>Civitai.com shows prompts and settings on almost all images on the site
The only exception are images generated offsite that don't embed prompt metadata, or videos because there's no equivalent EXIF block to store them in.
That's the reason for catbox links becase CB doesn't strip EXIF the way 4chan does (that's a story for another time).
https://files.catbox.moe/o3tkyd.jpeg
>>
Are there any AI video uncroppers yet?
>>
>>8973181
>uncroppers
I mean technically yes, since the Wizard of Oz playing inside the Vegas Sphere uses outpainting, but I don't think that's a thing for Stable Diffusion yet. If I'm wrong, someone correct me.
>>
>>8973030
It's not that long of a story. 4chan does it because people were hiding contraband and viruses in it, or doxxing people with picture's location data. 4chan strips all metadata from all files as a safety feature. It forgot to do it for PDF uploads because that feature was using someone else's prewritten code (ghostscript 2012), and look where that got us 4 months ago.
Catbox, meanwhile, is full caveat emptor.

>>8973181
like an outpainter but video? that seems doable. surely there must be one somewhere.
>>
>>8973192
>>8973207
interesting, thanks!
>>
File: ComfyUI_00569_.png (1.32 MB, 752x1504)
1.32 MB
1.32 MB PNG
>>
>>8973192
>I mean, there's nothing STOPPING you from using animatediff with a controlnet for outpainting, same as you might use it for inpainting to remove clothes etc.

I've just never seen it done.
>>
used to have Loras on CivitAI for real people like Laura Bailey etc. but privacyfags got it removed and most of my stuff got lost when I got a new PC recently

is there any repository where similar Loras are still available?
>>
File: flux_krea_00025_.png (1.3 MB, 800x1408)
1.3 MB
1.3 MB PNG
>>8973249
woops, wrong image
>>
>>
>>
>>
>>
File: 00003-2675838767.png (517 KB, 512x1024)
517 KB
517 KB PNG
>>8973343
How do you get two people into an image?
I've wondered about this. You make an image with one person and then edit in the other?
what is this process called?
>>
>>8973545
Are you not able to generate an image with two people through prompting alone? What model/service are you using?
>>
>>8973545
prompt "man, woman, sex"
it's that easy
>>
>>
>>8971338
this looks really good can you share the catbox? I'd like to see how it was made
>>
File: 00006-2471991372.png (3.1 MB, 1248x1824)
3.1 MB
3.1 MB PNG
>>8973568
I found a tutorial. I guess the model makes a difference. I thought the Pony model was just for My Little Pony pictures.. but I thought wrong.
Really ought to be better descriptions on the models.
>>8973575
>prompt "man, woman, sex"
If you have the right model...
>>
>>8973684
nice one!
>>
>>8973684
Pony was trained on a mix of Danbooru, e621, Rule34 and Derpibooru. So about a quarter of it is MLP. It responds well to tags from those image boorus, for example "flat_chest, small_breasts, medium_breasts, large_breasts, huge_breasts, gigantic_breasts" is a clear progression, without the ambiguity of natural language.

Only problem is, very little of the dataset was realism. People pushing it back to the photo look either make it lose much if its knowledge, or they don't go far enough.
>>
>>8973695
isn't everyone on illustrious models now? or is that old news
>>
>>8973698
It's noobAI now, Ilustrious is old and busted. But they both have issues with photorealism even worse than Pony,
>>
>>8973760
I thought noobai was illustrious. there's lots of models on civit that have both labels.
>>
>>8973771
Civit's labeling in insufficient here.

The Illustrious label on Civitai was originally used for version 0.1, which later forked into illustrious 1.0 and noobAI, by different teams. By the end of it noob was a big enough finetune to get its own label. They trained a lot more on Danbooru, added furry stuff from e621 and near the end experimented with v-prediction, which helps with lightning and especially dark scenes. You don't see that in realism tho as everyone seem to pick up the earlier eps-prediction versions for merging.

The illustrious branch went on to become a failure, 1.0 is still on Civit but not that widely used and afaik doesn't get many realism merges (if any at all), 2.0 is exclusive to some other site and 3.5 is paid online-only. They continued training only on Danbooru, yet somehow the model's knowledge ended up worse than noob even in anime-exclusive tags. And they experimented with higher resolutions, generating in 1536x1536 base res, but it didn't work very well.
>>
File: 00008-745185331.png (1.23 MB, 832x1216)
1.23 MB
1.23 MB PNG
>>8973693
they all tend to be the same!
You put in Missionary and you get this. You put in cowgirl and you get the same thing but now she's sitting on him. Doggy? Doesn't know that. kinda a one hit wonder...
>>8973695
seems like the words it was trained on are not natural language as I was led to believe, which explains my poor results. You have to know the magic words and what's in and what isn't in the model.
>>
>>8973929
Good thing there's a convenient wiki for all the magic words https://danbooru.donmai.us/wiki_pages/tag_group%3Asexual_positions
Or at least the ones Pony would have known before the realism conversion, some knowledge is always lost along the way.
>>
so that should be doggystyle, sex from behind
and also from_behind, if it's pov

assuming none of those loras and embeddings interfere
>>
>>8973930
that looks useful, thank you.
>>
File: 00048-1195443394.png (1.33 MB, 1024x1024)
1.33 MB
1.33 MB PNG
>>8973930
Yes! It works. Thanks!
>>
>>8973335
Can you show the school track and field teams at practice? or even at a meet?
>>
>>8973822
So the way forward if you want to train photorealistic without going Flux is NoobAI?
>>
>>8971217
What the fuck are you talking about dude?
>>
>>
File: civit training.jpg (40 KB, 1073x658)
40 KB
40 KB JPG
>>8973822
Then what is the deal with civit lora training not listing it as an option?
and also what are some good noobai models I should download to get started with then?
>>
is there a way to quickly generate a " simple" version of your image to just see what a promt does or how things will be positioned?
>>
File: test.jpg (956 KB, 1500x3451)
956 KB
956 KB JPG
>>8974354
Well, if you are set on using an anime sdxl model as your base then it should be noobAI now, not illustrious or pony.

>>8974452
I'm switching between SamBase 1.0-alt https://civitai.com/models/1602016/sam-base?modelVersionId=1823930and NAILove https://civitai.com/models/1252962?modelVersionId=1582207 (left and right). Sometimes based on the prompt, or base gen with one then upscale with the other, etc. And still on the lookout for others. Since such checkpoints are always a compromise between realism and knowledge lost, and each one's "realism" also has a different look to it.
>>
>>8974529
scuffed the link
SamBase 1.0-alt https://civitai.com/models/1602016/sam-base?modelVersionId=1823930
>>
>>8974507
to see if a concept is understood, just use a lower resolution and fewer steps. positioning is harder to reproduce, no guarantees. but if your quick-and-dirty test throws up a position you like, you can enhance and resize the result using img2img (as long as the proportions stay the same).
>>
>>8974529
Can you post just the top right image, for scientific purposes
>>
>>8974626
It was just a quick test gen, but sure. Hope you can read comfy noodles. https://litter.catbox.moe/yma8n8aqqkhtq5ec.png
>>
>>8974638
thank you kindly
>>
>>
>>8974390
1) Lustify XXL is on Civitai, but you need to give them 50,000 buzz for the next 11 days.
1000 buzz cost $1, so it cost $50, in 12 days it will be a free download.
https://civitai.com/models/573152/lustify-sdxl-nsfw-checkpoint
2) Not really into anal. So, for me, no anal is a feature.
>>
>>8974694
That the model was lobotomized is not a positive. It's not like you'd have gotten anal by accident without prompting for it, but who knows what other basic stuff is missing.
>>
File: 1.jpg (147 KB, 832x1216)
147 KB
147 KB JPG
>>
some of the people releasing these models seem to not really know what they're talking about. they'll give advice about generating with their models that makes no sense or is just obviously wrong. still a black box mystery for some of them even if their models can produce nice images.
>>
>>8974945
Lobotomized? You can't have everything. You know it's not even IN most other models, which is why as a newbie I tried getting it to work when I shouldn't have expected it to work. I had that epiphiny the other day, making stable diffusion images is a lot of work.
Someone has to train it.
>>
File: 20250908060240-4126807243.jpg (895 KB, 2432x1664)
895 KB
895 KB JPG
>>8975203
If you're using an Illustrious-based model, Illustrious was originally an anime model trained on the entire Danbooru dataset. It should know how to depict anal sex in an anime style, but if model bakers want to retrain it for photorealism, they should make sure that its original knowledge doesn't get lost.
>>
>>8975088
I would say most of them don't know how to use their own checkpoint. They recommend three lines of negatives that clearly don't do anything in an XY test, same with quality tags, and then the prompt looks like it was AI-generated. "A rich tapestry of deep red and orange, featuring swirling patterns that add depth and movement to the scene. Evokes a feeling of mystery." and then it spits out a regular forest cause the model has no idea. Or they try to use the Danbooru tag approach but don't check the tag wiki and just make up random "tags" on the spot.
>>
File: 00000-2599179944.jpg (499 KB, 1536x2560)
499 KB
499 KB JPG
finally got it working again
>>
File: bed.anna wawrzyniak.jpg (454 KB, 2160x1478)
454 KB
454 KB JPG
>>
File: witch.anna wawrzyniak.jpg (319 KB, 2160x1478)
319 KB
319 KB JPG
>>
>>8971220
>natural looking saggy tits
catbox?
>>
File: ComfyUI_00005_.png (1.22 MB, 896x1184)
1.22 MB
1.22 MB PNG
>>8976541
This is Chroma's tits style when you push for max amateur realism. I unfortunately seem to have lost that specific image, but I have this other one that should have the same prompt.
https://files.catbox.moe/08bpu4.png
>>
File: ComfyUI_00088_.png (1.61 MB, 896x1184)
1.61 MB
1.61 MB PNG
>>
>>
i prefer my slop on the barely probable side of realistic.
>>
>>8976862
That's hot as fuck
>>
>>
>>8976497
wow thats cute and sexy
>>
>>
>>8977591
Nice.
>>
>>
>>
>>8978380
>>8977591
Definitely more of this! Can you gen two girls showing off together?
>>
>>8978386
>>
>>
File: Workflow_video004.mp4 (210 KB, 288x512)
210 KB
210 KB MP4
>>8973314

You messed up when you gave me this
>>
>>8978619
kek nice
>>
>>8978398
They should be even hairier!
>>
Hows this?
>>
File: 20250912053743-2113201554.jpg (484 KB, 1664x2432)
484 KB
484 KB JPG
It's OK, it could use more saturation or a higher CFG scale.
>>
>>8978443
>>8978398
>>8978380
Love the hairy pits and pussy, looks almost real
>>
>>8979211
Interesting. A bit too much abs, but the atmosphere is right for the show.
>>
File: ComfyUI_15349_.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>>
File: 20250912064948-2948666425.jpg (384 KB, 1664x2432)
384 KB
384 KB JPG
>>
>>8979748
Love the "wings".
>>
>>8979974
Thanks. It was img2img+controlnet off an anime pic >>>/e/3000137, that and/or prompting her name must've helped cause the model didn't struggle with them at all.
>>
>>8979748
Box?
>>
>>8980009
Not sure it'll help, I tend to mess around a lot with different checkpoints before settling on a look. Could have been more straightforward.

Grabbed the image from >>8980008 and ran this upscale img2img+controlnet over it https://litter.catbox.moe/in8ro709xe4nyfy3.png, masked out the face for controlnet to ignore so she wouldn't end up looking like a huge-eyed anime person.
Then another img2img at much lower denoise and different checkpoint, to get a nicer face and less airbrushed look. https://litter.catbox.moe/8jvsb1u2lk2qz98j.png
And then some inpainting for the zipper, eye color, clutter on the table, etc.
>>
File: A&P final.jpg (798 KB, 2432x1664)
798 KB
798 KB JPG
"A&P" by John Updike. For a long while, I've tried using various realistic models to generate this, but now I realize that only Illustrious-based models understand the concept of "strap slip".
>>
>>8980028
Interesting, thank you! I mostly work within Krita, but will experiment.
>>
best model for realistic degen nsfw?
>>
>>8980056
I guess Sam Base is the best we've got right now, but it's so fried that its instructions on Civitai require you to lower the CFG scale to 1.5. The other thing that annoys me about these realistic checkpoints is that they give every female character the same nose.
>>
>>8980076
thanks, I am so behind on nsfw genning that I'm using cyberillustrious
>>
>>
>>
>>
>>8980056
The realism space is a mess, I'm using one single model for all of cartoons, anime and 3D, yet more than a half dozen for realism.

Sambase (at least the older one, 1.0-alt) kept the most Danbooru knowledge both in concepts and characters, but it's on the edge of what you could still call realistic. JuggernautXL is on the other end where it only takes natural language prompts, and most checkpoints seem somewhere on the spectrum between these two, but with their own bias in faces, lighting, color tone, etc.
>>
>>8980144
Do you do any showing their assholes?
>>
>>8975638
very true IMO there is a lot of slack to pick up in the negative prompt game and there are way too many padding and nonresponsive tokens in some of the sample prompts
>>
File: 00046-787545874.png (1.85 MB, 896x1216)
1.85 MB
1.85 MB PNG
>>
>>8980056
honestly at this point chroma, with the quant quality restorer added if you need booru tags
>>
>>8980682
>the quant quality restorer added if you need booru tags
explain what this is pls
>>
>>8980412
>>
>>8980900
>>
>>
File: 1736098190296006.mp4 (2.07 MB, 864x1024)
2.07 MB
2.07 MB MP4
>>8976565

Those breasts look amazing

>>8980900

mmm pancake mix
>>
>>
>>8980969
Very nice clip
>>
File: anna wawrzyniak.jpg (485 KB, 1314x1920)
485 KB
485 KB JPG
>>
>>8980904
>>8980900
i realize i had fucked up my prompts and generation parameters
>>
>>8981420
>>
>>8981003
Oh my god more like this, please!!
>>
>>8981003
This one is really good
>>
>>8981721
>>
>>
>>8981779
Keep them coming, this is so great!
>>
>>8981779
artist?
>>
>>8982092
I am ! Enjoy
>>
>>8982116
that is some prime elf pussy
>>
>>8981895
Demonic eyes.
>>
>>8982005
what he said
>>
>>8981420
>>8981436
Prompt?
>>
>>8982341
it's not dall-e
>>
>>8980890
Just google it.
>>
>>8980079
>cyberillustrious
people still use that?
>>
>>8982379
Not so much anymore.
>>
>>8980076
huge vixen pornosite vibes from this one
>>
>>8982386
It's because Vixen shoots everything with low exposures.
>>
why the fuck can't I use babes by yogi on civitai anymore?
>>
>>8982116
Hairy asshole elves give me life
>>
>>
File: 20250915034946-189396322.jpg (539 KB, 2432x1664)
539 KB
539 KB JPG
>>
File: 20250915064320-1339468789.jpg (473 KB, 2432x1664)
473 KB
473 KB JPG
>>
File: 20250914_1717_video.mp4 (2.04 MB, 1280x824)
2.04 MB
2.04 MB MP4
>>8982601

It's getting smarter xD

>>8980969

these are from back when we had multiple i2v slop threads.

>>8982727

animate this with the original as the start frame.
>>
>>8982863
>animate this
planning to eventually after getting enough pics to mass gen. those massive tits WILL be shaken.
>>
File: ComfyUI_00003_.png (1.23 MB, 896x1184)
1.23 MB
1.23 MB PNG
>>
Chantal from the accounting send you a pic
>>
>>8982938
Can she be an elf too? I want to see so many more of them
>>
File: 00000-980208452.png (3.03 MB, 1920x1072)
3.03 MB
3.03 MB PNG
FINALLY got img2img working right, turns out I had to make sure the DynamicThresholding settings were the same as for txt2img.

I've been building a harem. Here's Scarlett Sinclair, her daddy was an oil baron but her body belongs to me. I like to bring her to the park and have her model lingerie for me, it humiliates her deeply but you wouldn't know with how wet her pussy gets.

Getting the right prompt and dimensions for multiple views/panels took a while, and I've given up on adding cum or a buttplug, even though she'd look great in one. Just too inconsistent. But now I know I can just go into MSPaint and zoom in wherever I want before passing it to img2img, easily getting whatever close-ups I want from the main composition.

Real women are obsolete.
>>
>>8983112
>Real women are obsolete
*pictures of real women are obsolete
and soon videos too
>>
>>8983069
I'll do tomorrow
>>
>>8982379
haven't found another model yet that's good at futa, animals, milking machines, just crazy degen stuff
>>
>>8983282
illustrious is in the name
you want something based on noobAI, which is illustrious plus e621 knowledge as well as far more training on danbooru
>>
>>8983333
>noobAI
so a noobai realism model?
>>
>>8983335
yeah
>>
>>8983336
thanks bucko
>>
No matter what the fuck I do I cannot get a WAN2.2 workflow to work. Every single one is a fucking spaghetti nightmare of 1000 nodes and it never fucking works.

Even simple ones it fucking falls over at some point or just outputs garbage.
Even using custom scripts to install a complete ComfyUI environment with everything you need just doesn't work.

Why isnt there an easy way to set this shit up.
A1111/Forge is easy as piss, ComfyUI makes me want to tear my eyes out
>>
>>
>>8980076
Can you gen a 27 year old black haired college boy with a swimmers body stepping up out of a pool wearing nothing but a transparent black thong?
>>
>>8983918
No, are you the same person who asked before?
>>
>>8983916
These are amazing! Can you gen two office elves together showing off as well?
>>
>>8983923
Nude in the office?
More please.
>>
>>8983927
you are fin gourmet
>>
>>8983963
fantastic gen
>>
>>
>>8983967
Damn, righty got all the hair
>>
>>
>>8983983
She peeing into a bowl? That's a lot of ass hair, needs to be on her ring!
>>
>>8983931
>>
>>8983931
>>
>>8982727
another one I'm going to animate later
>>
>>8982888
>>8984245

excellent
>>
No one:
Me: make em clap
>>
>>8986271
and the world was a better place
>>
>>8986271
>>8986497
I'd love if they had hairy assholes!
>>
>>8983990
Now can you show a glasses wearing nude woman in the middle of a board meeting calmly pointing to the pie-chart with the others at the meeting just accepting it as normal?



[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.