File: 00001-2574630366.jpg (1.94 MB, 2304x3840)

1.94 MB JPG

>>8411832
oh hey one of my old 1.5 gens

File: cuta7seq8joc1.jpg (42 KB, 640x800)

42 KB JPG

And that was the recomended archival dump.

>>8411844
I do an archive dump when starting a thread to prevent cullings.

File: 030124 (t20).jpg (1.67 MB, 2048x3072)

1.67 MB JPG

yeah five months ago

File: 00185-490844986.jpg (1.53 MB, 2048x3072)

1.53 MB JPG

>>8411848
no I get it, just glad the tech has improved, everything from back then looks so plastic now

>>8411852
This tech moved fast. Thanks forever to the genious working on refining it.

File: 00012-1199947862.png (2.15 MB, 1152x1536)

2.15 MB PNG

i made a style lora and i'm fairly please with the results

File: IMG_0885.jpg (878 KB, 3131x1672)

878 KB JPG

cute!!!

What model should I use for that hyper realistic shit?

File: milf.jpg (811 KB, 2376x1968)

811 KB JPG

>>8412592

I like the gens of vendo pony realistic 1.2, I usually do all of my upscaling and inpainting through fooocus.

I'm struggling finding information about how to merge models with advance settings.
Is automatic1111 the wrong gui for this? Is there any extension to have more control over the merge?
I think I'm supposed to use controlnet with a detection model? But I cant find how can I use it for the control of the model merging.
Basically what I want to do is to transfer nsfw content like poses, penetration concept, nudity etc to a regular sfw model.
I managed to do this somewhat with just regular merging but the results are just too random and you have to fight with the prompt a lot.
I still don't completely understand how the "add difference" merging works or when to do it...

>>8412662
It's not going to work
You are producing a new model M = A - B + C
To make it work you would need a model A with NSFW concepts (you have that), but also need a model B that was the base for A before it had NSFW added
Only then could you isolate the NSFW part and combine it by adding it to C (the target new SFW model)

>>8412672
So there's nothing I can do about it?
Is there a certain way or order to merge models to improve the results of what I'm trying to do?
Like weight merging really small amounts (0.5 - 1) of several other nsfw models and then do a final merge of 50% with my original model, or use my original sfw model as a start point, then merge small amounts of tons of nsfw models to it.
How do I use the "add difference" properly?
I've had some problems/errors while trying to merge some models, which I find odd because I have a 4090.
Any tips?

File: 67uy678uiyk7i68.png (1.8 MB, 1024x1024)

1.8 MB PNG

cyberrealismxl is pretty good, most photorealistic models on civit are comparable, I see more results by experimenting with different prompts and samplers

>>8412653
Really nice one !

>>8413560
Prompt?

File: 00032-2523602265.png (1.31 MB, 1024x1024)

1.31 MB PNG

>>8413626
>amateur closeup iPhone photo of naked sweaty dirty grimy 30 year old Caucasian woman wearing steampunk toolbelt and looking away, brunette hair, natural saggy breasts, inside small cramped steampunk airship engine room interior, large steampunk dieselpunk iron machine, foggy, steam, extremely-detailed photo, hazy lighting, skin details, crisp clear detailed image
I make a base 3d render in blender and then img2img that, I hop around and try different samplers and stuff for each image

File: img-5ddDPNdg1yp9vskZKcADl.jpg (271 KB, 1152x1440)

271 KB JPG

https://getimg.ai/ref/OWX1gjisK

File: img-895tthRg4DS8gEUxEJzBR.jpg (277 KB, 1440x1152)

277 KB JPG

Made here:

https://getimg.ai/ref/OWX1gjisK

File: 1706268489147540.png (1.36 MB, 768x1280)

1.36 MB PNG

>>8399204
first try at controlnet

File: 1690854141484518.png (2.41 MB, 1032x1728)

2.41 MB PNG

feedback?

File: 00081-3024589506.png (2.17 MB, 1184x1728)

2.17 MB PNG

>>8414090
I like where your head's at.

Is SDXL worth using for realistic porn yet? Did it have its PonyXL moment?

>>8414495
1.5 is better than XL, pony is best

File: 00025-1258206524.png (1.19 MB, 1024x1024)

1.19 MB PNG

xl takes forever to get right but I personally like it for prompt adherence

>>8414495
i use models from sdxl and pony. both have hot models and give me great satisfaction

>>8414639
>>8414652
Which XL (non-pony) model has good prompt adherence?

>try all recommended samplers for ponyrealism
>all of them are either too slow or give me artifacts
>try euler a
>it just werks

>>8414090
Nice. The biggest issue is the face covering looking painted on the face.

>>8415185
Lower the CFG, it interacts with the sampler
The more "advanced" samplers take bigger steps so move faster into artifact territory
Of course Euler works, but it's slower and needs more steps for the same result that a better sampler gets

>>8415226
>Lower the CFG
I always use 7, IIRC 6-7 is recommended, so it should be something else, unless this recommendation is bullshit

>>8415235
that's your problem, most of the newer ones want more like cfg 4

File: Capture.png (37 KB, 994x702)

37 KB PNG

>>8415237
I meant this shit in the model's description

>>8415271
model authors are almost always dogshit at prompting so you should use that only as a starting point to test

>>8415235
I bet 95% of the people that use SD, including model authors, don't know what a sampler even does
It's just a solver for a high-dimensional differential equation
To find the right CFG for your specific use case, run an X/Y/Z plot with different values
Even if something doesn't produce artifacts, you might prefer a different value for artistic reasons
Also there are extensions that mess with the scaling

>>8414639
can you make this same pic but with emilia clarke?

>>8415271
does anyone see a difference between all these dpm?

>>8415584
DPM++ is state of the art currently
The subflavors differ in behavior of course (ancestral vs. SDE vs. non-stochastic ones)

>>8415612
What's the difference between them, in simple terms? Is, say, DPM++ 3M SDE Karras better than DPM++ 2M SDE Karras?

>>8415646
Samplers: https://stable-diffusion-art.com/samplers/
2M vs. 3M: https://medium.com/@furkangozukara/sampling-method-dpm-2m-sde-vs-dpm-3m-sde-with-schedule-type-uniform-vs-karras-vs-exponential-09ab0ac379dd

>>8415646
Difference is minuscule in practice
Given any fixed seed, switching out the sampler might produce a better or worse image always
That's because it's always a random process
The samplers use different heuristics to speed up finding solutions (images)
3M takes one more step per iteration than 2M, but that's no good if the gradient leads it right into a shitty valley

Greetings from 2022. Two questions.
First, what's up with all the big images? Last time I tried, I was only able to generate 512x768 max before I ran out of memory. Did everyone get better GPUs, or has the code improved?
Secondly, I remember every generation looking simple and plastic like this >>8412842 (sorry anon, heh), but now we have generations indistinguishable from real life, like this one >>8397903. What happened, and is there a guide on how to achieve this affect?
Sorry for essentially asking to be spoonfed with a cement truck. I'm just surprised.

>>8416181
What GPU are you using?

>>8416185
1650 Ti

>>8416188
Well, I generate 1024x1024 on my laptop with 3050 without any issues, but 1650Ti is an ancient card by this point, so you might want to get something better.

>>8415104
lustify or biglust is the best non-Pony that I've seen

>>8416181
what happened is SDXL and then Pony Realism— they pretty much brought a whole other level of coherence to gens, it's still nowhere near coherence levels of bing, but hey

and yeah, people have good gpus for upscaling and whatnot

>>8416191
>>8416210
Dangit. Guess I'll have to stay with tiny gens for now. Thanks anons.

File: 00006-265937458.png (1.14 MB, 1024x1024)

1.14 MB PNG

>>8416181
yeah upscaling and inpainting goes a long way, best way to get high quality gens is with img2img, even just a rough sketch so the AI isn't starting from scratch

Was directed here by /g/ /ldg/, copying my post.

Good SDXL models for softcore NSFW? I'm training a realistic concept model (girls peeing, yes I know I'm weird) and I'm trying to find the best base model / mix to train on. I have ~2000 well tagged images. The model needs to be able to do lewd, know what a pussy looks like, good at poses / anatomy, and ideally be pretty flexible / not overfit. Some things I've tried or will try soon:

Base SDXL: nope, mediocre anatomy, and lack of pretrained NSFW makes it shit for fine tuning on my dataset.
Juggernaut: not good enough at NSFW. Seems weirdly overfitted too. Training on this gives terrible results.
Halcyon: similar to juggernaut in limitations, but the 1girls look really good IMO.
Iniverse: pretty good actually.
bigASP: extremely flexible base model, probably best results of all I've tried. A bit difficult to wrangle, need to play with quality tags a lot.
Aphodite: haven't trained on it yet but it does softcore NSFW way better than most of the popular realistic models.
Anteros XXXL: haven't even downloaded it yet but looks promising. It's just SDXL trained on 1M+ photos (similar to bigASP in that regard) so might work really well.
Literally all hardcore models: way too fucking overfit on hardcore porn. Anteros might be the same way actually, we'll see.
Pony realism models: not realistic enough, too big of a style shift when training on my dataset. Results always come out bad. Maybe skill issue on my part somehow.

Anything I've completely missed that meets my requirements?

>>8416206
Thanks, will try those

>>8416324
Lustify was just mentioned
Biglust is a merge of lustify and bigASP apparently
There is also irisLux
From my experience it feels like it was trained on brazzers style porn
I don't think it prompts too well, but it seems to understand anatomy
So maybe it's not so overtrained and fits the bill actually

Which juggernaut did you try? The new X is completely different from v8

>>8416341
thanks fren I will test these out

I tried Juggernaut X. Is that better or worse than v8?

>>8416351
I would say v8 is lewder
In general I found that it's always a good idea to test an older version of a checkpoint
Often they overhaul their approach completely when releasing a new version, so then it's a totally new model basically

File: 1721608365207644.png (1.05 MB, 768x1280)

1.05 MB PNG

>>8416296
>>8416210
>yeah upscaling and inpainting goes a long way, best way to get high quality gens is with img2img
What's the QRD on upscaling? I thought everyone was just using highres fix since that's what's making my gens look better for some reason and all images posted in these generals are smaller than 2000px
>best way to get high quality gens is with img2img
what's the difference from the extras tab upscaling?

>>8416446
different method, you will gen new details with img2img with your model, extras tab is usually just some algo and not a several B parameter model

File: 1722382470.png (43 KB, 817x1270)

43 KB PNG

>>8416078
oh hey, it's the generic pony version of a favorite

>>8416341
Okay I've done some gens, both Lustify and BigLust seem really good. Will try training on both of them. Lustify feels like a generic realistic model that does NSFW really well. And BigLust is just bigASP but the people look a lot better.

Also did some xyz grid comparisons of some other models. Aphrodite is kinda shit, bad prompt comprehension, anatomy worse than competitors. Anteros XXXL has promise though. The creator also released two different pre-aesthetic-tuning checkpoints of the model, which are more "raw". Might be interesting to train a lora on one of those then merge it with the final model. A lot of my images are not super great quality, so this might sidestep the quality degradation problem that happens to some extent when training.

>>8416969
fuck anon what model?

>>8416983
BigLust

File: 00001-991280697.png (1.27 MB, 1024x1024)

1.27 MB PNG

>>8416446
I sometimes use adetailer which is basically automated inpainting, but by img2img I mean I usually start with an image first, I use a 3d render but you can use anything, a photobash or whatever, also good to use a higher res image and then let it scale it down to 1024 or whatever, even with a really high denoising strength it will give you a better composition imo

File: 1695305577388352.png (16 KB, 351x285)

16 KB PNG

>>8417056
thanks a lot anon. Just to be clear should I be using SD upscale script? Ican't select an upscaler otherwise.
>also good to use a higher res image and then let it scale it down to 1024 or whatever
I did but I set the resolution here and I think I fucked up because it took 40 minutes and froze at 100% lol.
>but by img2img I mean I usually start with an image first, I use a 3d render but you can use anything, a photobash or whatever
to get the composition and poses? you don't use txt2img at all?

>>8416324
>>8416341
if you're using bigasp, make sure you're using these tags:
https://gist.github.com/fpgaminer/e835328c7883b1929419273f6e73c3aa

File: 00052-4109155863.png (1.18 MB, 768x1280)

1.18 MB PNG

I think this is my first attempt at something at least marginally tasteful

>>8417056
moar gravyard shit that's hot

>>8417277
Is there anything like this for AnterosXXXL?
I like its style, but have no idea how to prompt it

>>8417481
I think that model was trained on full sentences, so it does whatever the fuck it wants

>>8417509
Yeah it's not danbooru tagged for sure, the author would have mentioned that anyways
But there is no guidance at all, so I guess it boils down to whatever captioner he used then

>>8417434
Looks good.

File: 00034-713529581.png (1.06 MB, 768x1280)

1.06 MB PNG

I'm using pony realism as a base and trying to use goddess of realism as a refiner but I keep getting these prominent white dots in the image. does anyone have any idea as to whats causing this?

>>8417612
What happens when you change the refiner switch to lower or higher?
And with lower CFG?

File: IMG_0893.jpg (639 KB, 1212x1974)

639 KB JPG

I so badly need a img2vid service that’s uncensored lol.


Does anyone know how you could lower the denoise for inpainting in fooocus? I know you can do it easy in A1111, I was trying to get some cum all over chadchad’s lower face but couldn’t find how to lower the strength to paint it on :(

>>8417661
I adjusted when the refiner switches and even my step count. Going to try cfg later.

>>8417885
It's under the Advanced/Debug tab

>>8417538
Looking through the example image prompts, it seems like it was trained on forum post comments
>This woman gives the best head. She is practically swallowing the man's dick. What a great blowjob!
Or scraped from porn ads

> bigASP was trained on about 1,440,000 photos, all with resolutions larger than their respective aspect ratio bucket. Each image is about 1MB on disk, making the database about 1TB per million images.
the first really serious attempt in a like a year. he rented 8xH100s. that dude is based. too bad the tagging process had unpredictable effects.

>>8418253
How much it would cost to train something like that?

>>8418258
He ran it for 5 days, so north of $3600.

What lighting-related keywords work in ponyrealism?

>>8418384
https://danbooru.donmai.us/wiki_pages/tag_group%3Alighting

those should, I'd think

>>8418462
nta but thank you for this tag group link

>>8418253
https://www.reddit.com/r/StableDiffusion/comments/1egwgfk/joycaption_free_open_uncensored_vlm_early/

He's working on a bigASP v2, with its own custom captioning model. If it goes well and is a big improvement over v1, this could end up basically being the Pony of realism models.

>>8418911
How would that compare to helloworldxl which was trained on a dataset tagged using chatgpt4?

File: 00002-563992674.png (917 KB, 1024x1024)

917 KB PNG

>>8417218
yeah sd upscale script is good, that will fill in the image with more detail as it upscales
I don't usually use txt2img, I find it best to start off with an image and then work from there, at least to give the AI a rough idea of what I want

File: BigLust_00181_.png (2.63 MB, 1576x1080)

2.63 MB PNG

>>8418911
This is sick.

>>8419054
HelloWorldXL is trained on only 20,000 images. It's barely more than an aesthetics tune of SDXL. bigASP v1 was well over a million, and the creator said it seems undertrained still. Also the tagger used for v1 as well as this new captioner supposedly work much better for NSFW, whereas GPT-4 will just not describe certain things.

>>8419087
Holy shit

File: 24214125005-bba144.jpg (404 KB, 768x1344)

404 KB JPG

Non-pony porn models are a breath of fresh air
Getting tired of the pony look that you can never get rid of
And of course fully celeb compatible

>>8419492
Which ones are you using?

File: 24213014450-498842.jpg (376 KB, 832x1216)

376 KB JPG

>>8419496
Playing with anteros-xxxl for now
It certainly does glamor as well as really amateur looking pics
But prompting it is an adventure

>>8419811
Nah, it's too way late to ban anything

>>8419811
They use the civil courts to circumvent free speech
But they won't get names from 4chan posts, so...

File: IMG_0895.jpg (360 KB, 1212x1974)

360 KB JPG

I <3 MILFS! :)

>>8419492
Pony prompting is just too easy, I feel too spoiled to go back. I tried BigLust and BigASP and it was miserable for me.

File: file.png (1.51 MB, 768x1280)

1.51 MB PNG

New base model dropped
https://blog.fal.ai/flux-the-largest-open-sourced-text2img-model-now-available-on-fal/

>>8419998
Tried the example prompt with the bearded man in SD 1.5
My image actually looked better
>wew

>>8419998
>new model that can be run locally, and has good coherence
YES YES
>24GB ram gpu

ah...

>>8419998
>another online service
no thanks!

>>8419890
this. pony is really fun and easy. also, pony has already many fetish, without loras required

>>8419071
nice
>>8419075
so many themes, but you only like the boring blacked theme...

>>8420172
For realistic stuff, a regular photorealistic model will always perform better though (but prompting will be limited due to captioner shortcomings in NSFW)
Pony is only good for tentacle stuff, gaping holes, giantess/size difference, monster tits/dicks etc.
For good bondage stuff for example you already need LoRAs anyways, even for pony

I think I'm gonna try more img2img
Doesn't matter if you can't prompt stuff in that case

>>8419075
Post tight sluts like piper perri if you can

>>8420642
The future of hot tub streaming

>>8420155
Its local you stupid

>>8420179
said the nogen

those are great anon what model are you using?

File: 00087-3775993592.png (1.44 MB, 1024x1024)

1.44 MB PNG

>>8420731
I usually use stanleys triweekend mix, or epicrealism, I start with a 3d render and then img2img that, then I sometimes inpaint and photoshop the results

>>8420687
kys faggot. nigger cocksucker

>>8420746
>, I start with a 3d render and then img2img that
oh you already answered me before sorry lol. I started using img2img and its great for getting the composition and poses but my gens are still a bit mangled, I'll need to learn it. Is a 3d render better than an image?

any grifters here? I'm seeing accounts with really large followings just posting the same characters with the same dead eyes and seven fingers, most of us here are better than that, how to get in on that action?

>>8420797
you mean they are passing them off as real women on onlyfans?

File: BigLust_00218_.png (2.59 MB, 1080x1576)

2.59 MB PNG

>>8420748
kek do it yourself

File: GT01DgqWsAAuglv.jpg (334 KB, 1680x2160)

334 KB JPG

>>8420853
aye
here's one, I guess it's not super obvious if you're not used to seeing it but the hands are pretty fucked imo, maybe they just block people who call them out on it, and the people who don't engage just don't get recommended that account anymore

>>8420904
well I can tell you this much I'll do it as soon as I figure out how to get the same face on different gens

File: 00725.jpg (1.56 MB, 3296x3840)

1.56 MB JPG

>>8420797
I'm not, but I've been putting some deep thought into starting. Especially since the jews in charge are looking to pass laws to restrict it.

File: 00728.jpg (1.54 MB, 3296x3840)

1.54 MB JPG

>>8421222
Whoops, I meant "our respected, and dignified elected representatives"

>>8420904
It's kind of painful to look through every detail of an image and fix it though
Would be way easier to work with an "ugly" girl and inpaint her to make her "pretty", with complete face swap of course

>>8420906
Use a LoRA or embedding
When you use a different checkpoint, celeb faces can become unrecognizable but still look attractive
Also it's really easy to train your own LoRA
But you also need consistent background in your gens, and clothes
Your own LoRAs can do that too, in fact stuff like that gets picked up from the training data by accident usually

>>8420902
Where did you say you were genning these?

>>8421505
You are in an SD thread, people run this shit locally

File: 1699771124061764.png (184 KB, 695x769)

184 KB PNG

>>8421330
but you need a real face to train the lora on right? If you're going to grift coomers pretending they're real girls it's best to avoid legal repercussions by using real people, unless you use an obscure estonian instagram user or something.
I've got consistent faces just by locking the seed before, idk if it works on all models but I guess that's an option too to create a woman out of thin air?
>It's kind of painful to look through every detail of an image and fix it though
I don't even coom to my gens cause I'm always fixing imperfections and adjusting settings to get what I want. It's tragic really. But seriously normies and especially boomers don't have an eye for this stuff at all, they'll see the worst multi fingered chinese brainrot AI image and swear it's real. If you can somehow get into the boomer market it's a gold mine.

>>8421646
>If you're going to grift coomers pretending they're real girls it's best to avoid legal repercussions by using real people, unless you use an obscure estonian instagram user or something.
Just use photos of some random girl from a 3rd world country, she'll never know.

>>8421646
>>8421651
You can use some unnamed stockphoto model, change the hairstyle with inpaint and they'll never prove it's her
Can also make some rando gens where you describe age, ethnicity, body
Then feed those few images into LoRA training
Make more gens with the first LoRA, then train another
There are guides for making character LoRAs from even a single image only
Or mix two celeb LoRAs
There are "rando person" LoRAs on civit also, for extra safety

>>8416969
Best image in the thread, and excellent taste.

>>8421545
I run Anteros XXXL and Helloworld with celeb loras via a site on a phone. Most of my results look way better than the stuff here, so no not everyone is stocking piling 50 gigs and buying more ram for it.

>>8421915
>Most of my results look way better than the stuff here
How about you post some then?

File: 00005-3398433625.jpg (501 KB, 1536x2560)

501 KB JPG

I wish I knew why genning an image with ComfyUI seems to use more VRAM than a near-equivalent operation in Forge.

File: succ_0192.jpg (771 KB, 2432x1664)

771 KB JPG

I feel like the pony realism models do a better job at realistic exotic fantasy women than you can get from 1.5 versions, though they still give off something of a B-movie costume vibe. Maybe inevitable

File: 00071-4260296451.png (2.76 MB, 1568x2016)

2.76 MB PNG

every monitor in SD is running Windows 10

>>8421915
Is it free? Or are you paying someone for that privilege?
RAM is cheap, what about GPUs and electricity?
If you want to run any other checkpoints, do inpainting, img2img, then what do you do?
Feel free to post some samples, celeb posting is allowed

File: 00053-2430128497.png (2.91 MB, 1456x2016)

2.91 MB PNG

>>8421966
this is a rabbit hole I didn't even think existed, thanks anon!

File: 546tryh.jpg (160 KB, 1024x1024)

160 KB JPG

been messing around with flux and it gets some interesting results

>>8422000
As it should be.

File: BigLust_00516_.png (3.32 MB, 1536x1536)

3.32 MB PNG

>>8422175
>>8422174
Looks nice so far.

File: 1716050711292322.png (1.33 MB, 1024x1024)

1.33 MB PNG

>>8421966
>>8421943
>>8421940
that's just base ponyRealism?
>though they still give off something of a B-movie costume vibe. Maybe inevitable
add some freaky prompts

File: succ_0141.jpg (492 KB, 1536x2560)

492 KB JPG

>>8422392
>>8422143
Nice ones!
> that's just base ponyRealism?
Yes ponyrealism v21 with some inpainting. Sometimes I use a workflow that lets me inpaint the faces with another model but it is tricky, especially with unusual skin tones

File: image-17.jpg (231 KB, 1024x1024)

231 KB JPG

some more flux experiments

>>8411828
I've got a question that's gonna be posted in the other thread too

Anyone here has tried CIVITAI or NAI? Which one is better (in your opinion)? SD moves too slow on my 2000 box

File: 1718736154539421.png (1.16 MB, 1024x1024)

1.16 MB PNG

is /asdg/ the slowest of all SD threads?

File: tron.jpg (213 KB, 1120x1440)

213 KB JPG

>>8422510
yes

File: 1697229316318092.png (3.91 MB, 2048x2048)

3.91 MB PNG

>>8422409
anon can you spoonfeed me a good upscale workflow?

>>8421915
What's your favorite anteros prompt? There isn't a lot of guidance or example images

>>8422175
>>8422463
>>8422467

Prompt : Refugees welcome

>>8422613
Nice, prompt and models ?

>>8422972
>>8422973
>A room with a view.

File: succ_0237.jpg (630 KB, 1536x2560)

630 KB JPG

>>8422613
I usually take the base generation, then manually fix details in GIMP, then use ultimate sd upscale and inpaint details manually after that.

File: 1718385366525793.png (1.24 MB, 1024x1024)

1.24 MB PNG

>>8422967
https://files.catbox.moe/5u9lll.png

>>8422971
What model are you using to get such a nice face?

File: 1722655308831057.webm (1.26 MB, 1152x896)

1.26 MB WEBM

live update from the special needs dall-e thread who managed to make a good post after months of trying.

>>8422971
Mad Max? Bravo sir

What's the best model for realistic looking people but mainly, realistic versions of fake people? I saw a crazy good Ellie someone made but every Time I try and create something like it my pics always look super "splotchy"

I'm trying out SDXXXL and was wonder if anything better existed?

Sorry for the text wall I'm new at this and need some help

>>8424489
if you post the Ellie image I can tell you what they used.

File: 1722725449139494.jpg (105 KB, 832x1216)

105 KB JPG

>>8424497
thx, there were a couple

File: 1722257004349727.jpg (119 KB, 1216x832)

119 KB JPG

this was the other one. didn't wanna just request it since idk if thats allowed

>>8423586
cute

>>8422561
More tron pron!

>>8424502
lol, i wasn't expecting you to post them.
the first one seems like a 1.5 model, or img2img using a 1.5 model, and there's dozens of models on civitAI you could use.
the second one looks like a generic XL girl, and there's no loras posted for Ellie on regular realistic XL models.
you could look at Lustify/natural beauty/true to life/biglust.

>>8423344
>https://files.catbox.moe/5u9lll.png

Thank you !

But how can I get the prompt and the model/lora used?

>>8424607
the lighting is kinda different so the head may just be inpainted on

File: eti-topkek-kakaolu-35-gr-(...).jpg (550 KB, 900x900)

550 KB JPG

>>8424587

File: D06790D6-EDCB-4581-A334-A(...).jpg (398 KB, 896x1152)

398 KB JPG

captcha S0X0S

>>8419075
how about some cute blondes or redheads >>8411832
?...

File: 20240804143958_00001_.png (2.64 MB, 1168x1600)

2.64 MB PNG

>>8425280
Blondes it is.

>>8424489
Zonkey 5.0 with (real life:1.5) in your prompt is the best pony realism model that I've used.

Otherwise BigLust is another good option, not as great for specific characters or celebs though

>>8424504
this could also be multiple models in a comfy workflow

File: 20240804170655_00001_.png (1.11 MB, 872x1000)

1.11 MB PNG

>>8425380
What the F. This Model (and the Prompt real life:1.5) is insanely good!
First Creation from my standard workflow with this model loaded in.

File: 20240804174950_00001_.png (2.26 MB, 1168x1600)

2.26 MB PNG

Fuckin Hell

>>8425131
download that image and put it into "PGN info"

So how hard is SD on a GPU?

Hi Emma dalle lora anon, did you share your Billie model too? The images i saw you post so far look great! Thanks

>>8425536
>https://files.catbox.moe/5u9lll.png

Ah great thanks!!!

File: test.jpg (616 KB, 1362x2334)

616 KB JPG

>>8422368

Need a little bit of help with prompting for proper skin texture, but BigLust is pretty neat. slutty teens for all!

>>8425800
My tests from prompting "realistic skin texture" suggests that the base model was trained on cg textures with that label
Some models show realistic skin when prompting for imperfections, skin wrinkles, folds, but might result in older looking people
"highly detailed", "sharp focus" etc. can help, but it's checkpoint specific
Haven't tried "airbrushed" or something in the negatives
Sharpening/noise shaping extensions can help, but it's finicky
If your checkpoint is trained with glamor porn shots, you might just be out of luck since it's just fine tuned for airbrushed looking skin

>>8425883
I also noticed that hi.rez fix, img2img upscale and adetailer all bring a strong airbrush compared to the original gen

>>8425865
This is really good

File: 00027-1495506522.jpg (1.98 MB, 1344x1728)

1.98 MB JPG

>>8425898
glad someone enjoys my gigacoomer plastic skinned babes

File: 00013-2638945618.png (1.01 MB, 768x1280)

1.01 MB PNG

Damn, ponyrealism is pretty good at selfies. This bathroom looks like shit tho.

>>8425865
more

>>8425959
yeah i wouldn't be paying the tiler!

>>8425941
What model is this? Looks amazing

>>8426077
ponyRealism_v21MainVAE

>>8426098
Thanks anon

File: 20240805001229_00001_.png (1.82 MB, 1120x1376)

1.82 MB PNG

Tried something there.

>>8426152
I wanna rub my cock against that belly

File: 20240805011121_00001_.png (2.11 MB, 1120x1376)

2.11 MB PNG

i'm a Virgin, i swear....

THERE HASNT BEEN EMILIA CLARKE POST IN A VERY VERY LONG TIME.....

File: 00000-1482246963.jpg (1.93 MB, 1344x1728)

1.93 MB JPG

>>8426152
my mercy came out incredibly thick

>>8426306
go ahead then

File: 20240805013243_00001_.png (2.11 MB, 1120x1376)

2.11 MB PNG

>>8426311
she is really great! care to share your workflow? i'm especially interested in how you get it so damn crisp looking.

>>8426306
that's because she is having fun somewhere else ;)

File: 24218012535-e17db2.jpg (413 KB, 768x1344)

413 KB JPG

>>8425897
Very realistic gen
>>8425941
Plastic faces, but the bodies are super oiled up (with reflections!) and would pass, great stuff!
Also great hands, but dry looking (separate inpaint?)
>>8425959
For non-pony XL, I noticed that beauty tags and the common negatives affect bathroom cleanliness
>>8426214
>>8426216
Not at all. The only good thing about her death was new LoRAs coming out
Only did some quick tests with her so far

I've seen people using local LLMs to create prompts from a vague ask, does someone have a guide for that for comfy?
Can you force that on a second gpu?

>>8426332
Huh, why were those Shellys deleted? Did I miss something?

>>8426334
Wouldn't need a separate GPU since you can just run the prompt expansion before the gen(s)
Fooocus does that when the Fooocus V2 style is activated, using GPT-2
Probably not so useful for SD actually, but practically required for FLUX

>>8426323
>you get it so damn crisp looking.
Went to bed but the secret is sending the image in the img2img tab upscale x1.5 or 2 at noise 3.3 +/-.3 and use adetailer for the face (use custom seize or it will fuck up the face)

>>8426347
I was more checking something like the latest llamas, gpt2 is quite old.
But that's the idea yeah, hopefully I can find a node that can do that.

>>8426360
You should ask in /g/ldg/ where they are playing with FLUX in comfy
That thing needs really descriptive prompts and I saw someone mention prompt expansion with LLM the other day

>>8426372
I'll look in the archives then, thanks anon.

File: 1722825706.png (1.47 MB, 1024x1536)

1.47 MB PNG

>>8424502
i don't even remember making these lol

>>8426736
did you make em lol? if so could you share the model and catbox?

File: 00139-1380918876.png (1.03 MB, 832x1224)

1.03 MB PNG

they are some of the best ive ever seen!

pic related is a realistc attempt at maya

File: lazymix.png (800 KB, 640x960)

800 KB PNG

>>8426754
i don't remember, it may have been this
> https://litter.catbox.moe/dytjdx.safetensors
or one of the civitai 1.5 loras with good preview images.
here's the negative embedding i always used
> https://litter.catbox.moe/e4yq6m.pt

the model was one of the non asian ones, like lazymixrealamateur or one of uber models. you can even try one of the furry mixes like "everythingfurry".
the easiest thing would be to make a lora of her for one of the regular XL models tho.

File: 00092-2546506944.png (1.02 MB, 832x1256)

1.02 MB PNG

>>8426830
Thanks! any idea what the prompt was? ive only been genning for like a week now so sorry for all the questions lol

>>8426830
opened the file raw and i think its from this for anyone else intrested https://civitai com/models/131493/ellie-tlou?modelVersionId=144529

Are the highest quality celebs on Civitai? I found a few of the Dall-e ones that were made, but it doesn't seem to be very effective.

>>8426839
something like
> 1girl or woman, from behind or from side, looking back, nude, red shirt
background depends on the model, some will just ignore your prompt.

>>8411844
Catbox?

File: 00015-2580665176.png (953 KB, 1024x1024)

953 KB PNG

>>8426332
>>8426337
Post-nut clarity hit hard, anon.
Funnily enough I made those two pics before she died and never got to post them. I always thought she was hot.

File: 00020-2717160252.png (1.11 MB, 832x1216)

1.11 MB PNG

>>8411828
>futa
/futasdg/ has been made for realistic and cartoon styled futa
>>8427183

>>8427169
Shelley Duvall was so beautiful.

>>8421915
This is such a strange LARP

>>8427254
ah yes I totally made that up for uh.. what exactly? your (you) a day later? lmao

>>8427256
>Doesn't share a single gen or prompt
>Replies within 5 minutes of being called a fraud
Totally legit

File: C131CC66-E01F-4B3E-AFE4-5(...).jpg (574 KB, 768x1344)

574 KB JPG

>>8427267

I said I gen them on my phone.It's almost as if I could also notifications on my phone to replies or something crazy futuristic like that.

>>8427341
>>8427354
Not bad gens, but you were shit talking about better results than what is posted here, using anteros and celeb LoRAs
Rando pinups aren't gonna cut it, post moar
The thread is in bump limit anyways, no one will complain about spamming

I’m having trouble getting two specific characters to not start swapping traits like hair and eye color when I use Forge Couple. Do some Lora just refuse to play nice with each other? I had no such issues when I tried it with other characters.

Also, are there any good Lora for Pony that you can set one character to be only slightly shorter than another? I’m talking one characters is about shoulder height to another. All the height modifiers that I’ve seen on CAI make them into midgets.

>>8427391
Bleed between subjects is common
It even happens for single characters between different parts of an outfit, background/furniture etc.
You have to use regional prompter or inpainting to get what you want

>>8427398
Oh boy, I was afraid of something like that. Inpainting is a rabbit hole that I’ve been avoiding.

>>8427360
Well I don't post celebrity porn for the obvious possible future legal ramifications of our gloriously corrupt legal system. Those are from Flux though, not the AnterosXXL site/app, but it seems to be better quality just without the ability to easily do celebs. you can, but it's a pain and lots of luck.

>>8427468
What are some good anteros prompts, or other tips?

>>8427468
They can sue the site then and get your phone fagging IP that way
If you are scared of the legal system, gen locally and never share gens
Don't use a service, fucking lol
That's the best way to get caught

New thread

>>8427624
>>8427624
>>8427624