Thread 8920752

Previously, on /asdg/ >>8901996
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

>>8920756

Hey, Vasquez! You ever been mistaken for a man?

>>8920756
wood nut in cider

I hate what this is doing to me, I'm reading dozens of pages of guides and shit. I'm even downloading feet Loras so my gens are more holistic. I just wanted to fap guys.

>>8920764
pls prompt

>>8921363
What kind of shit checkpoint requires feet loras

but otherwise yeah, I know what you mean. Gets worse once you start tinkering with the gimmick techs like regional prompter or ipadapter, or baking your own loras.

>>8920475
Slightly improved pic after changing denoise.

>>8922024
prompt?

I fucking love this technology. But sometimes, I can't get it to obey simple directives like FACING FORWARD / STRAIGHT ON view. This is the ONE good gen out of like 20 ... do you guys see anything wrong with my prompt?

https://files catbox moe/95wl9s.png

>>8922651
Sometimes it just happens. Either use controlnet or up the weight on position tokens.

>>8922651
You're using bigLust but prompting it with tags like an anime model.

>>8920757
stunning, please teach me

>4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
From the OP link I found
https://sprites.neocities.org/metadata/viewer
This is useful for like ComfyUI or Forge rite? And is as powerful as the inspect link for NovelAI

>>8923055
Looks like a CartoonPink LoRA on a realistic model.

>>8923076
>CartoonPink LoRA
Think I got it, thanks. Wish I knew his prompts for tits like that, will test it out tho but I'm just starting out. Thanks again.

Thanks for notifying me of the "sheer" tag. Occasional spatial continuity errors, but very pretty.

getting sd to run with amd feels like it's infinitely harder this year compared to last one. it's to the point where i'm at the point of giving up and settling with rendering with my processor instead.

>>8924201
So all the blabber about finally progress being made on AMD usability was just fucking smoke?

>>8924281
i'd say. can't speak for how the latest cards are, but as someone with an rdna 2 card, it's still shitty to get running since all the linux guides are years old, and don't work with modern firmware.

the only one that sorta works is sd next, but even that's super wonky, and takes 5-15 minutes to get a decent result. previous pic unrelated.

>>8924291
Bummer.

>>8924384
Previous pics were satisfactory txt2img where even with no other loras besides the clover clamps one, it came out decent. I then ran this thru img2img but change the checkpoint to Pony as I cannot gen from scratch a txt2img with Pony. It seems more realistic and less glossy compared to Illustrious MMMix. I've got a few more checkpoints downloaded but I personally prefer more time to familiarize and get satisfactory gens out.
Maybe Pony is more friendly for non bdsm stuff (or needs good Loras and excellent prompts) as in the past some wizards in Realifying threads had trouble a bit with my request, but one of them did share their setup and thats how I first learnt about Forge and Pony.

>>8924388
Illustrious has a better prompt interpretation than Pony.

>>8924473
and noobAi has better prompt comprehension than illustrious

Just bought an Nvida GPU, I was using AMD/Zluda to get SD running before.

What would anons recommend for UI, Is Swarm a good place to go after screwing around with 1111 for a while? Also should I use Stability Matrix to play with multiply models?

>>8924475
>NoobAI
First time finding out about these, thanks. So these checkpoints are good to start? https://civitai.com/search/models?baseModel=NoobAI&modelType=Checkpoint&sortBy=models_v9&query=noobai

And like NoobaiRealistic, DucHaiten-NoobAI and DucHaiten-NoobAI-Cinematic might be suitable for /asdg/?

>DucHaiten-NoobAI
scratching this off. Its cinematic one seems more suitable by seeing their examples.

>>8924509
there's too many variants now, I can't keep track
but basically noobAI is branched off illustrious 0.1 and trained further by a different team

>>8924480
just get reforge ui if you intend to just use sdxl models.

>>8924519
there are better model i recommend over duchaiten models.
creativitij https://civitai.com/models/1398870/creativitij?modelVersionId=1581191
cyberrealistic xl
https://civitai.com/models/312530?modelVersionId=1962475
pornmaster pro realism
https://civitai.com/models/1045588/pornmaster-pro-illustrious-and-noob
https://civitai.com/models/1045588/pornmaster-pro-illustrious-and-noob
realdream sdxlpony17
https://civitai.com/models/153568/real-dream?modelVersionId=1906612

>>8924724
He was asking for models based on noobAI

>>8924509
I use SamBase https://civitai.com/models/1602016?modelVersionId=1812943
but I care more about getting my anime girl cosplays accurate than about a 100% photorealistic look

Has anyone figured out getting sde samplers to work on chroma?

depressing how slow these threads are. Even is gif video threads are faster than asdg.

>>8924724
>>8924738
Thanks. Would check them soon. Its ok, for NoobAI an anon mentioned about improved prompt interpretation,
>and noobAi has better prompt comprehension than illustrious
and Illustrious MMMix so far can flexibly produce what I fairly want, characters, misc stuff (restraints, objects) only unless there were too new (kpop demon hunters xD). I was quite used to MMMix which to me is like 50% of either 3D looking and less photorealistic here for this thread, and I appreciate the Illustrious suggestions as well.

Last pair for today, between mmmix and its Pony conversion.

>>8924827
Most of my stuff is either too experimental or just a "not-anime girl in a bikini" which I guess is too tame.

i like gaslighting the AI

post something, these threads are dry.

holy based
https://civitai.com/models/1822855

>>8924961
I wanted to complain about him tagging it black_shirt but not sure anymore. Is that some extreme v-neck with a raised collar, and just white skin underneath? Or is it a white shirt and a black jacket on top.

>>8921856
Right now I'm learning what an Adetailer is, I need to make my faces better. This pic was a failure, but still made me laugh. I wasn't expecting crazy eyes over there.

>>8924771
Very nice, what model and detailer? I like that the reflections match more or less.

>>8925535
>>8925537
this is the photorealism general, by the way

>>8925538
oh shit, my bad.

>these threads are dry.
I kind of envy the MMAI thread btw xD
>page 10
A spare from my gens a couple of days ago, but just wanted to thread bump.

Genning some pics upon trying out SamBase. Actually my laptop is kinda basic or 'potato' for this, and also doing this casually. I'm not sure if there is any other better Lora for clover clamps. This one is for Illustrious but the page it can be used for NoobAi too https://civitai.com/models/1115059?modelVersionId=1253025. I got a Joschek Lora that works so excellently for Pony, but I've to work around for Pony if my pics were bdsm (preferably just img2img attempts). As the civitai page cautioned about the chain of the clover clamps, it can be good or really random, and each of my pic gens take several minutes each time, pushing my basic system.

>threads are dry
Hope no one is in weekend mood already haha, so maybe I would try to enliven this thread xD

>>8927164
Btw, I'm new to this checkpoint and not sure if I overspammed the prompts but I would like prompt each hand to be in different poses. Like maybe I was using danbooru tags like when using MMMix previously.

>1girl, nude, beach, day, hot, absurdly long green hair, seiza upright, relaxed, medium breasts, big blue eyes, cute face, nervous, smile, sweat on face, (right_hand_on_own_thigh, left_arm_behind_head), bikini on floor nipple clamps, nipples, clover nipple clamps, silver nipple clamps, nipple clamps chain

>>8927164
Looks a little burnt, try lowering your CFG value to like 3-4. You're also using "score_" negatives from Pony, and high weighted "quality" negatives from way back in SD1.5. Can also just prompt "photorealistic" in positives instead of trying to guess which form of "cartoon, anime, illustration" actually does something. General advice: negatives are a bandaid for specific issues, not something to rely on for every image. If you have more than a few words in there or have things you never remove, that means your positive prompt could be improved instead.

The "imperfect eyes" come from her being far away, if the face is too small AI can't make details properly. Can solve this automatically with Adetailer, or by inpainting the face afterwards. Both of those do a localized crop+upscale, to give the model more resolution to work with.

>>8925538
this makes me scared. Was the MMMix model acceptable in here, or is /csdg/ meant for checkpoints that have cartoon lines/borders and anime/hentai types, especially if there are border and lines around the model?

>>8927175
All of danbooru's hand tags should work: hand on own hip, hands behind head, hands at sides, squeezing own breast, peace sign, clenched fist, whatever. It's not perfectly reliable but just prompting two conflicting hand poses/actions should give you one of each.

>>8927176
Felt curious as this was my usual prompts without knowing actually what exactly they were.
How "score" helped for Pony, and for cfg is it like to lower its brightness?

Can any-anon do something classy with Audrey Hepburn?

Found out that she was an actual authentic war hero in addition to being an actress.

Nudity or no nudity but keep it classy please.

Heck will give points to anything with Elizabeth Taylor and Hedey Lamar as well.

[No I said HEDEY Lamar not HEDLEY Lamar]

>>8927183
As you can see even the hentai pics are still up, so there's no "danger" if you cross the line besides maybe annoying some people. It would be impossible to draw a clear line anyway when it's more like a spectrum from 3d to photo.

>>8927219
score_9, score_8_up, etc. were quality tags back on Pony. And I don't think they did much at all on photorealistic pony variants, but most people just copy others without thinking. It's very easy to run the same prompt and seed only changing one tag, see what effect it actually has.

>for cfg is it like to lower its brightness
Not brightness, contrast. Something's deepfrying your image, her skin is red with black spots everywhere and the sand looks like a carpet. I thought it might be CFG or the excessive negatives but after some tests that doesn't seem to be it. Maybe the clamps lora. Pic is what the model outputs for me.

I went back to the SamBase page: https://civitai.com/models/1602016?modelVersionId=1823930
>How run:
cfg 1-5, step 8-20, sempler euler_a, sheduler beta.

Yeah I was genning with the old prompts and was using the default settings of Forge, including cfg, sampler and scheduler that I didn't change.

Forge default settings had like steps 20, cfg 7, sampling method: DPM++2M, scheduler: Auto

So this is kind of vastly improved after seeing that one line recommendation in checkpoint page.
>sempler euler_a, sheduler beta.
Cfg is 4, and lowered sampling steps.
So each checkpoint I've got to follow any instructions they've put in their civitai page. But anyway thanks for pointing out my first pic, yeah its definitely messed up using the incompatible settings on a different checkpoint.

Leaves me wondering further, what is DPM++2M and Euler_A and scheduler Automatic vs Beta?

Audrey

I can't believe I was so far genning MMMix with sampling method DPM++2M previously instead of Euler or Euler A. It only works for Pony v9 that I was previously using and I didn't change the default settings in Forge not knowing the difference earlier XD

If MMMix is no issue here, sometime later I would be back with those, but I'm also excited with SamBase for now and the Euler A switch.

>>8927362
>So each checkpoint I've got to follow any instructions they've put in their civitai page
Or experiment. Not every model baker knows how to best use his own model.

>>8927362
>Leaves me wondering further, what is DPM++2M and Euler_A and scheduler Automatic vs Beta?
So first off "Automatic" just means "use Karras for DPM++ sampler variants, and Normal for everything else".

It's much easier to see in cartoon styles, but in general Euler tends to smooth things out while DPM is more noisy, which is supposed to be better for realism unless you're going for an airbrushed look. Not sure why it's causing major issues with this checkpoint.

CFG is basically a slider going from blurry and grey to over-sharp and high contrast. The SDXL base model can in theory handle CFG values of up to 20, but the more training is done on it the worse it looks at high values. Regular photorealistic models can often still go above 10. But stuff based on Pony/illustrious/noobAI is basically SDXL re-trained into anime, only to be re-trained again back into realism. Because of that the usable range is far lower.

Hold me I'm falling

>>8927362
The Sampler stuff doesn't matter as much as you'd think, outside of working vs not working. Anything ending with _a stands for "ancestral" which basically means that the image will not converge on a "solution" but will continue changing with higher steps (not necessarily to the better). Non-ancestral samplers will converge on a "solution" image. Automatic vs beta is germane past works/glitches.

Lowering sampling steps will decrease your time but I've never had luck with my setup at values <18. There are models that work at very low step counts but I've never used them.

As stated above, CFG is roughly analogous to contrast. It's more than that, and higher CFGs give you more colorful images, but that's 99% of what it is.

>>8926965
>>8926966
love Rin

Initially when I was genning simple and few poses so they were like glamour model poses the checkpoint was ok for me. But along the way I started not getting good gens. Not sure you could get the girl's features smooth or young like. When I put expressions like fear, scared, shocked the faces don't really look good. It makes her wrinkly compared to what I intended. I only have just very few proper ones. Also I've been spending hours but cannot get good gens of the stationary pillory, but maybe its too ambitious of an idea. So far at best the checkpoint gens out her being tied to a plank instead of a proper pillory.
For this pic I don't get why she has 6 fingers despite my best efforts.

....

>>8928658
These 2 are bad ones, in terms of expressions.

Even trying to put a Lora of stationary pillory worsens it, so this rope tied to the pillory is actually better. I think an anon mentioned something about dataset, so a previous checkpoint could make a decent one though not perfect, and I think I overstepped this checkpoint and its meant for vanilla stuff ultimately.

>>8928660
First one a bit odd, but.the second one isn´t hat bad.

>>8928664
Models/Checkpoints are tend to be generalist unless it´s a finetune.
That´s why there are a trillion LoRAs for almost everything under the sun.

>>8928668
Yeah, lol. I don't know why, there are 6 fingers, but I wanted to depict her being shy/ashamed. And the hair was "strawberry blonde" ended up including strawberries. I thought that is a character's attribute based on danbooru.
>the second one isn´t hat bad.
MMMix but I think I got the term 'semi realistic' checkpoint, it did successfully gen out a proper stationary pillory like in >>8926830 but I could only get her rope tied to a pillory.

>>8928668
Yeah, lol. I don't know why, there are 6 fingers, but I wanted to depict her being shy/ashamed. And the hair was "strawberry blonde" ended up including strawberries. I thought that is a character's attribute based on danbooru.
>the second one isn´t hat bad.
Maybe as it is more realistic, but adding expressions to me makes the model look older and the expressions were wrinkly.

>>8928681
One of the problems with female faces. The single line being out of place and the whole face just looks wrong.

Your Fellowship roll was really good.

Bump

How do you upscale your stuff? I've been using cyberillustrious and the base gen looks good with a fair amount of detail with Euler, Normal but when I upscale it a lot gets smudged and everything just looks too smooth
I upscale with NMKD Superscale Sp 17800 and then I resample with a tile controlnet and Euler, Normal at 25 steps and 0.33 denoise

>>8930780
I suggest you ask in /g/ for a quick answer.
Sorely lacking knowledge in that area.

>>8930780
Just the old RealESRGAN_x2plus. Maybe run the upscaler on its own, see what it outputs and if that's already smoothed out. But I doubt it'd make much difference even if you just used Lanczos, at 0.3 denoise the AI's gonna change it anyway.
>tile controlnet and Euler, Normal at 25 steps and 0.33 denoise
More or less the same, Euler A instead of regular but that shouldn't matter; xinsir's promax union instead of tile and with no preprocessor; and I adjust denoise between 0.15-0.5 depending on how the image responds.

>>8930835
I guess I just gotta a play a little more with denoise then
I changed from Normal to Karras though and it helped a little

1/8

2/8

3/8

4/8

5/8

>>8931305
add a train of black guys please

6/8

7/8

8/8

>>8931389
>>8931397
catbox?

If I want to train a LoRA using a dataset of real photos to gen with say, that SamBase model, do I train directly on it or on NoobAI?
I know you're supposed to train on the base model, but I don't know if that also applies for realistic models that are so far apart from the base.

>>8931945
If you have the time you should try both, but I would expect training directly on sambase to work better.

>>8927194
>>8927199
Unsubbed

https://files.catbox.moe/7k9ok1.jpg
https://files.catbox.moe/pw3nsr.jpg
https://files.catbox.moe/i2g38t.jpg
https://files.catbox.moe/3dobp6.jpg
https://files.catbox.moe/n6ambe.jpg

>>8927220
Here's an old one

>>8933374
another

>>8931945
How did it go?

Requesting:
Natalia Dyer from I Believe in Unicorns/S01-02 Stranger Things.
Winona Ryder from Lucas, Beetlejuice, and/or Alien: Resurrection.
Emily Kinney from The Walking Dead and Masters of Sex.

>>8920752 (OP)
Is there a name for this specific type of shirt, where it only has a tease of midriff? I love that look.

>>8933553
For prompting, I mean.

>>8933554
on illustrious models, you would prompt for midriff peek.

>>8933555
illu and noob
on anything else you might as well use a lora right away

After a while I think I found the nice settings for Sambase, also after checking back its civitai page
> cfgScale:3, steps:12. sampler: Euler a
Like even 4 was too bright for me but 3 was just nice, and moderate number of steps as my earlier gens looked like, too sharp not sure how else to describe it.

For this image is it Sambase? >>8927238
How does she have smoother and younger features? Any Loras needed, like characters? I haven't used any character Loras so the default model seem to produce like mature looking and I need to be careful with her expressions prompting.

>>8933526
I am doing my part.

>>8933677
A very nice job.

>>8933568
It is Sambase, I lost the upscale metadata but here's the raw if you can read Comfy noodles https://litter.catbox.moe/sr692htfyu28dphe.png
Just a quick test gen, no loras, no negs, not much of anything.

I come from /h/ where I learned to prompt in a minimalist fashion, make sure every word has a clear effect on the model. Figure out how your UI does XY charts and you can run some experiments yourself with just a few clicks, see how it responds to different prompts, negatives, sampler/scheduler/cfg, loras on/off, etc.

>moderate number of steps
First time I've heard of too many steps making things worse. Usually it's just overkill past a certain point and a waste of time.

>>8933949
Cute!

>>8933375
>>8933374
Turner Classic Movies with sex scenes added in
If we're going to hell let's go in style

>>8934247
that really good
nice work

>>8934253
Thanks anon, have another:

>>8934664
hey thanks!
she's more my type

>>8933988
Thanks anon. Just testing out and fairly new to Sambase, so experimenting around. I randomly have anime girls in mind but might come back with western cartoon girls as well.
So far without Loras, Sambase has its interpretation of characters and its interesting for myself to see their Sambase 3d type versions.

>>8934703
>>8934717
I think it's a cool look. The eyes are unsettling in a good way if that makes sense.

>>8934720
>The eyes
I think its a characteristic from the anime girl. In the series she has that unique sparkling eyes that looks like crosses. I was trying to replicate Shokuhou Misaki but it ended up as an effect lol.

>>8934727
Ah. Makes sense. I really like the result.

>>8934727
Could try doing a second pass over the face, either inpainting or Adetailer. I think they're just melted due to being too far away from the camera, not enough resolution to get the details right. When I tried it would just give her regular human eyes unless I prompted for them with "+ +".

Also they're freaky huge for a human head, but maybe that's intentional.

>>8934972
other than not getting her forelock in the hair this is amazing

>>8932901
what is the name of that character?

>>8935217
Danbooru tag is titania_(sao) in which she is Asuna Yuuki.

What are /asdg/'s favorite shitmixes?

>>8935762
Sambase is the new hotness.

>>8935762
I'm the original Sambase shill, cause I'm mostly doing anime stuff even in this aesthetic. Old Lustify helps in img2img or inpainting to get a more realistic look.

But that's not a recommendation anymore, just me getting tired of trying out new merges every week.

>>8912420
>>8912345
can you share the vibe transfer bundle? thanks

I wasn't aware and didn't know much about celebrity Loras getting scrubbed. Not the requester anon, but any idea what happened chronologically, and like I tried searching for Audrey Hepburn Loras but I don't know where to find them anymore.

>>8936338
Credit card processing companies, read Visa and Mastercard, turned their attention to civitai, the go-to site for LoRAs. Civitai caved in and decided to remove among other things celeb LoRAs (they were previously available but unusable on-site). Then came huggingface and tensor.art. Now the best bet is to hunt a gofile link in the celeb threads in /b/.

>>8936349
Or bake your own. Doesn't take much effort, some twenty pics and 15 minutes of GPU time.

>>8936305
>>8912420
Yeah, that is pretty great.
Share a catbox?

>>8936734
It's NAI, metadata is in there. Made with vibe transfer though, that is their variant of IPadapter. Won't do you much good without the source images too.

Since this fag >>8931945 never reported back, I tried it with a quick and dirty bake. Works better when trained directly on the realism merge (right pic).

https://www.imagefap.com/pictures/4108930/XOXO-Leah

1/5

2/5

3/5

4/5

5/5

>>8936624
ooga

Requesting -

-a nude telephone switchboard operator at her switchboard with her head set .

-a latina woman , about 25 , wearing enginers pants and a machinists cap but nude from the waist up and with a few light engine grease smears standing outside an open repair garage looking at the viewer smiling.

>>8936953
>>8936954
>>8936955
>>8936956
>>8936957
>Her face when she sees this.

>>8936871
4chan removes the metadata.

>>8937415
And yet there's a plugin that shows the metadata

>>8937415
nai also embeds the metadata in the image itself, not just the exif section

>>8937415
Guess I should've clarified that they're not my gens, and I read the vibe transfer thing out of the metadata that is clearly in there.

>>8924201
>>8924281
>>8924291
What are you talking about? I have an AMD 7800XT and Fedora and it worked out of the box, comfyUI automatically installs RocM. I didn't do anything special.

>>8936353
Do you have an easy to understand tutorial on that?

>>8924771
Requesting background prompt please

is inpainting not working on tensor.art right now?

A general question. I was attempting to gen kind of a set with different variants. The intention was to keep the same pose like how Danbooru has master and child posts with the same pose but only small changes, or some random artists in any platforms i.e. DA.

Each time I try using the same prompts and same seeds, the pose already changes by the next gen. I believe I read somewhere that making variants is really hard and the UI would just creatively change it. It is attempted txt2img, maintaining the same prompts (or just small additions) and using Forge. Perhaps it is too much of an ambition idea. If my goal wasn't variants, single pictures are satisfactory for me.

Any tips on how to maintain the pose or best go along with how the UI produce the pics? I have no idea yet about inpainting yet and how to start with that.

>>8939295
ControlNet. https://github.com/Mikubill/sd-webui-controlnet

1/4

2/4

3/4

4/4

I've been making storyboard txt2imgs in Forge, and I just got Jan and a NSFW LLM to auto-generate erotic prose. As with any serious application, I'll be editing the text generated (I'm a published author) and if it takes off I'll work on improving the images as well.

Two questions:
>Can I profit from this? e.g., on Twitter/X
>Wanna see an example?

>>8939382
Can you profit?
There are plenty of Patreons out there.
Find a more interesting checkpoint and hit hires.fix.

>>8939389
What other checkpoint can do this? I'm using
>pornmasterPro_noobV4.safetensors
with VAE
>sdxlVAEE_sdxlVAE.safetensors

I like it because I can create a character and do a whole storyboard in one shot.

>>8939393
Then the problem is the technique. It would be wiser to gen the individual images at the best size possible and then edit them together.

>>8939398
I'll mess around with rescaling. The tracking of outfits, appearance, etc. across panels is important for narrative continuity, so I want to keep the storyboard pipeline.

So what do I do to profit? Make an X account and start making threads, and link a Patreon?

>>8939382
Diff anon (but the bdsm clover clamps guy using Illustrious).
>Wanna see an example?
Seems interesting. Is it images together with some storyline? Like if its a lot of images you can upload a set to catbox, together with its narrative, as a sample?

A couple of days back I was just dumping here my own mini sets of progressive pics but no real story.

>>8939402
How to profit?
The method you mentioned seems the most common, but beware it will be an uphill battle anyway you try.
It´s a very crowded field, so better look for an angle, like a franchise or a paraphilia.

>>8939382
There's no money on Twitter/X. It's just a gateway to Patreon/Subscribestar/DeviantArt, where you sell subscription access to your pics.

An additional precaution is Patreon being vulnerable to piracy where it can be uploaded to Kemono. But Subscribestar is probably kemono-proof and DA too, you can sell subscription.

>>8939560
people have found workarounds for patreon too
sending passwords in DMs, invalidating the links every week, etc.
Lots of artists on these days have the full import on kemono yet no real content in there.

Hips

>>8928388
damn, great gen

>>8939795
nice
looking forward to spooky season gens

>>8939911
>>8939907
>>8939887
I need more vintage like this, prompt?

>>8939942
Prompt: Close-up candid 1950s photo of two 24-year-old housewives (facebook profile:0.7), innocent faces, wearing a blouse and no pants, pubic hair, standing on staircase landing, aesthetic, flickr 4k, ancient architecture, girlfriends
Negative prompt: oversaturated, overexposed, ugly, mutilated, disporportionate, extra limbs, extra fingers, deformed, disfigured, mangled, disorder, malformed limbs, fused fingers, twisted body, washed out, poorly textured, plasticized, render, matte, shiny, doll, (3d, 2d, drawing, ((anime)), (cartoon), child, penis), accident, camera, lopsided, walking, airbrushed, mangled, broken, faceless, amputee, disconnected, crowd, vertical chain, marching, squat, fat, uneven, small head, fake, implants, bimbo, ugly, long, underwear, pants, skirt, topless
Chroma, clip skip 2, etc.

if someone would put this into a fucking machines img2vid it would be much appreciated actually i can't find the lora

Im using FLUX on Comfyui. Just got it working today. Can do some softcore stuff and should have dl'd a nsfw model. Any advice to get it more unfiltered?

>>8939964
use chroma instead or start downloading loras

>>8939956
I'm currently still Civitai's bitch so no flux-tier models with negative prompts and they only NOW got Krea dev, but tyvm

>>8939967
Thanks I already been using lora.safetensors which is meant to be the nsfw one but its tame seeming...trying to dl pornmasterPro_noobV4.safetensors.

Really ive just been getting Grok to help me build all this, which is very handy.

>>8939367
Semirealistic Pony img2img

>>8939989
Why are there like a hundred pornmaster checkpoints now, it's all the same shit...
Anyway none of those are Flux-based afaik, just SDXL.

>>8939956
Excellent set.
>>8940558
It´s always like that when something becomes the flavor of the month.

>>8933949
>Marin Kitagawa gives us bros a Monday treat xD

>>8940673
that's quite a treat

>>8940917
Bolin (under)
Mako (right)
Tenzin (left)

>>8941122
How do you even remember their names, I just call them earthbro and firebro.

>>8940917
Good stuff. Bit of a shame how his glans melted upon touching her mouth, could have used an inpaint pass.

>>8939794
nice

>>8940662
thank you

>>8939987
is tensor too sketch for you or what? honestly though we should both just stop this shit is addictive and useless unless i can start selling these shits to freaks even sicker than myself

if anyone has tricks to getting bigger tits in chroma let me know. gigantomastia works but is all or nothing

>>8941409
some of the Loras work but i am against non-style Loras in general because of their effect on image quality

Any good prompt tips to make images more porn?

>>8941413
dicknipples in negatives

>>8941418
>>8941420
That's pretty creative. getting quaint looking women and splattering them with cum like that.

>>8940917
wow that's a good one

1/4

2/4

3/4

4/4

>>8939344
>ControlNet.
Sorry for the delay to reply, but this is appreciated. Do I rely on the ControlNet already built in within my Forge, or externally installing the Mikubill is more advisable?

There are also quite a lot of options. Let's say I wish to derive a set of variants from one good gen, usually the random first gen and I like the pose, which preprocessor do I choose? Any Model to select, next to preprocessor?

Didn't know this was here

>>8942285
They should be the same.
Haven´t been able to use it for a very long time (thanks daddy Google), but my best guess is reference for pre processor.

>>8942301
A very nice set.

In an experimental idea, I tried to manually (prompts) include both Jinu (in demon form) as well as Rumi. So far there is no Lora yet in Civitai for Demon Jinu, and my failed gens had the system mixing up their clothes, like gat_(hat) and even hanbok, korean_clothes plus other Jinu characteristics going to Rumi at times lol. Anyway I reduced my prompts so only Rumi is performing here, its a slop as no character Lora yet. I was thinking of having a scenario where demon Jinu would do something to Rumi but that would have to wait I guess....

Here is one accidental and unintentional gay gen lmfao

>>8942285
Preprocessor should match the model you're using. There are a couple options for models. Lineart (also called canny or edge) only sees edges, works better for cartoons and anime. Depth only sees depth, sucks with flat color drawings works better with 3D renders or ptohographs. Normal map deals with reflections, it's better for object geometry than for people. Openpose tries to estimate the bones of major limbs inside a person, sort of like the skeleton system 3D rendering programs use. Less strict in regards to bodyshape, but may fail if the person is not fully visible or otherwise hard to estimate his pose.

There's also xinsir's promax and promax union, which do a little bit of everything.
And for models based on Pony/illustrious/noobAi there's also cn-anytest, which is some combination of edge/depth/normal while mostly ignoring colors.

Others are utility controlnets, for tiled upscaling, inpainting, style transfer, etc.

I swapped to an external SSD and now loading models/first gen after switching is unbelievably slow. It was more or less instantaneous before this. Using ForgeUI. Any thoughts?

I tried manual prompting without Lora for Rumi. Sambase gave her horns lol
>masterpiece, simple background, night sky, concert, cowboy shot, 1girl, rumi (kpop demon hunters), teen, high braid, forelocks, single braid, very long purple hair, dutch braided ponytail, white sleeveless turtleneck, white shorts, big eyes, smiling, singing, holding microphone

>>8942611
either it's not an SSD or the external link is slow

>>8942679
Ty. Fixed it by playing with the ports. Not sure why it was f'd and showing
>Device can perform faster when connected to USB 3.0
My generic usbc to usbc cable works. That + tweaked some settings and its now back to how fast it was when it was on my Nvme.

>>8941409
turns out the flux-krea lora understands "gigantic breasts" as a trigger who woulda thunk

ai fuckups are pretty cool aesthetically imo and i lowkey think that in 50 years there'll be a sort of nostalgia for SD 1.4 glitch art if we're still around and still have nostalgia

>>8943096
>years there'll be a sort of nostalgia for SD 1.4 glitch art
I think it'll be similar to scrambled porn.

nubread?

>>8944194
Wait until we are on page 10.

>>8944204
nubread?

sssshhhh.. breadbakers are sleeping
let it die

Giving this up for a while and so I'm attaching my workflow. Hope 1 of u finds it useful

https://files.catbox.moe/yiy08h.png

>>8944234
would probably find you baking the new bread more useful ngl

>>8944236
Idk how to do that so idc sorry lol

>>8944237
ello el xDD

>>8944081
l'origine du monde

>>8944234
damn, another checkpoint to test

Baking.

New thread:
>>8944432
>>8944432
>>8944432

>>8944429
>>8944434
Thank you.

>>8944387
Try both pornmaster checkpoints desu, they're both great