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

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • There are 97 posters in this thread.

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


[Advertise on 4chan]


>>2757494 hit the image limit
Please keep loli on >>>/b/ and western styles on >>>/aco/asdg

> Where do I start?
https://rentry.org/sd-mashup

> What's with catbox?
4chan strips metadata containing prompt info. When someone asks for catbox, they want you to upload the original image with preserved metadata. Prompt information can be loaded from the PNG Info tab of the WebUI.
Catbox is free and doesn't require registration.
https://catbox.moe/
Also consider litterbox for temporary storage.
https://litterbox.catbox.moe/

> Models and Mixes
>>>/t/1190826
https://rentry.org/sdmodels
https://rentry.org/hdgrecipes
https://huggingface.co/WarriorMama777/OrangeMixs/tree/main/Models
https://huggingface.co/andite/anything-v4.0
https://huggingface.co/syaimu/7th_Layer

> LoRAs
https://gitgud.io/gayshit/makesomefuckingporn#lora-list
https://rentry.org/hdglorarepo
https://rentry.org/trashcollects

> Recommended Extensions
These can be found in the Extensions tab if you're using Automatic1111 WebUI.
sd-webui-controlnet
ultimate-upscale-for-automatic1111
a1111-sd-webui-tagcomplete
sd-dynamic-prompts

> Recommended models for /e/ content
Anything-V3.0, NovelAI, AbyssOrangeMix3 / AOM3-A1, nutmegmix, g-anonmix
7th Layer Anime, Anything-V4.5

> Resources
Bad prompt fix embed: https://huggingface.co/datasets/Nerfgun3/bad_prompt
Catbox 4chanx extension: https://gist.github.com/catboxanon/
Big index: https://pharmapsychotic.com/tools.html
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups
Highres fix calculator: https://preyx.github.io/sd-scale-calc/
LoRa Training guide: https://rentry.org/lora_train
/g/ anon's advanced merging tutorial: https://rentry.org/BlockMergeExplained
Model similarity calculator: https://huggingface.co/JosephusCheung/ASimilarityCalculatior

> Related
>>>/aco/asdg
>>>/d/ddg
>>>/g/sdg
>>>/h/hdg
>>>/trash/sdg
>>>/vg/aids
>>>/vt/vtai
>>
File: jessica.jpg (673 KB, 3072x1152)
673 KB
673 KB JPG
>>
File: shantae.jpg (505 KB, 3072x1152)
505 KB
505 KB JPG
>>
File: mario-princesses.jpg (405 KB, 2304x1152)
405 KB
405 KB JPG
>>
File: alita.jpg (1.22 MB, 3072x2304)
1.22 MB
1.22 MB JPG
>>
File: mina-ashido.jpg (471 KB, 3072x1152)
471 KB
471 KB JPG
>>
File: 00065-3066951596.png (1.76 MB, 1008x1408)
1.76 MB
1.76 MB PNG
>>
File: pervertexhibitionist.png (2.16 MB, 1024x2048)
2.16 MB
2.16 MB PNG
>>
File: 01546-1479058621.png (1.78 MB, 1016x1416)
1.78 MB
1.78 MB PNG
>>
File: 01566-1406187965.png (1.86 MB, 976x1416)
1.86 MB
1.86 MB PNG
>>
Are there request rules?
>>
File: 00069-244125645.png (338 KB, 512x768)
338 KB
338 KB PNG
Bless the LoRAs list and everyone who makes them.
https://files.catbox.moe/h3wgdj.png

>>2762104
Speaking about requests, i'd kill for a (game) Serena LoRA. Tried making one myself but i keep getting errors.
>>
>>2762104
OP here. This came up in the last thread but I forgot to add a blurb in the new OP. I'd like everyone's thoughts on the matter. Here's the blurb I'm considering adding:

This thread is generally not for requests. However, if you want to fulfill requests, make an announcement. If you want to make requests, check the thread first to see if anyone is fulfilling. All things in moderation, please don't turn this into a request thread.
>>
File: 00052-47478529.png (1.47 MB, 1024x1024)
1.47 MB
1.47 MB PNG
Any tips / loras on getting a character to face forwards? I keep giving small ass prompts which always results in a back / side view.
>>
>>2762128
Have you tried adding "facing away" and "from behind" in the negative? Perhaps even add "facing viewer" to the positive.
>>
>>2762129
Also consider instead of positively prompting for small ass, negatively prompt for "wide hips", "thick thighs", and "huge ass".
>>
File: 00039.png (3.41 MB, 2048x2048)
3.41 MB
3.41 MB PNG
Hi
>>
>>2762043
set up your vae properly, the colors are washed out
>>
File: 00249-288043414.png (1.61 MB, 1024x1024)
1.61 MB
1.61 MB PNG
>>2762129
>>2762130

Throwing in the negatives seems to have helped.
>>
>>2762011
>>2762047
>>2762050
catbox?
>>
File: file.png (1.32 MB, 960x1280)
1.32 MB
1.32 MB PNG
Nep!
>>
do edits done with sd go here to?
>>
File: 01588-3947195466.png (1.74 MB, 976x1416)
1.74 MB
1.74 MB PNG
>>2762203
https://files.catbox.moe/4skhdt.png
https://files.catbox.moe/7pzndt.png
https://files.catbox.moe/1kxffp.png
>>
>>2762213
thanks anon I can't find the "akira" and "sohn" lora, you have a link ?
>>
Can't find any ai to help to tune this into the style of ranma 1/2
>>
>>2762211
Sure. And if you don't mind, share with us how you're using it to get the results you want. SD is a tool for much more than img2img.
>>
File: 01570-2503476485.png (1.82 MB, 976x1416)
1.82 MB
1.82 MB PNG
>>2762215
here's the akira one
>>2761922

here's the sohn one:
https://files.catbox.moe/5g988b.safetensors
it might not do great with some models, but they work well enough
>>
File: 01632-474065298.png (1.76 MB, 976x1416)
1.76 MB
1.76 MB PNG
>>
File: 00067-4086143105.png (1.85 MB, 976x1416)
1.85 MB
1.85 MB PNG
>>
File: 00283-2331987188.png (569 KB, 680x680)
569 KB
569 KB PNG
>>2762003
boota
>>
File: 00041-4177323731.png (308 KB, 544x544)
308 KB
308 KB PNG
>>2762334
>>
File: 00567-441059876.png (280 KB, 544x544)
280 KB
280 KB PNG
>>2762335
>>
File: 00439-725975142.png (359 KB, 544x544)
359 KB
359 KB PNG
>>2762336
>>
File: 00442-3262594562.png (360 KB, 544x544)
360 KB
360 KB PNG
>>2762337
>>
File: 00390-993903779.png (387 KB, 544x544)
387 KB
387 KB PNG
>>2762338
>>
File: 00052-1770957236.png (337 KB, 616x616)
337 KB
337 KB PNG
>>2762339
>>
File: 00573-2119018576.png (487 KB, 512x728)
487 KB
487 KB PNG
>>2762340
>>
File: 00686-1521124494.png (509 KB, 656x656)
509 KB
509 KB PNG
>>2762341
>>
File: 00184-1867364378.png (696 KB, 680x680)
696 KB
696 KB PNG
>>2762342
>>
File: 00111-1524569282.png (451 KB, 616x616)
451 KB
451 KB PNG
>>2762343
no more boota
luv roxanne
simple as
>>
File: 00145-1524569316.png (434 KB, 616x616)
434 KB
434 KB PNG
>>2762344
>>
File: 00396-250580957.png (367 KB, 544x544)
367 KB
367 KB PNG
>>2762345
>>
File: 00304-4279199617.png (212 KB, 512x512)
212 KB
212 KB PNG
>>2762346
>>
File: 00459-2095729245.png (497 KB, 512x776)
497 KB
497 KB PNG
>>2762347
>>
File: 00919-545471335.png (485 KB, 512x728)
485 KB
485 KB PNG
>>2762348
>>
File: 00005-2515187926.png (381 KB, 616x616)
381 KB
381 KB PNG
>>2762349
>>
File: 00045-919921008.png (458 KB, 544x544)
458 KB
458 KB PNG
>>2762350
>>
File: 00236-435137973.png (268 KB, 512x512)
268 KB
268 KB PNG
>>2762351
>>
File: 00649-507221778.png (471 KB, 512x728)
471 KB
471 KB PNG
>>2762352
>>
File: 00214-3980466579.png (250 KB, 512x512)
250 KB
250 KB PNG
>>2762353
>>
File: 00661-862793148.png (324 KB, 544x544)
324 KB
324 KB PNG
>>2762354
>>
File: 00090-253116957.png (472 KB, 616x616)
472 KB
472 KB PNG
>>2762355
>>
File: 00337-1276456695.png (314 KB, 544x544)
314 KB
314 KB PNG
>>2762356
>>
File: 00290-2428657747.png (371 KB, 616x616)
371 KB
371 KB PNG
>>2762357
>>
File: 00040-2282860462.png (537 KB, 512x728)
537 KB
537 KB PNG
>>2762358
>>
File: 00048-3775902641.png (454 KB, 616x616)
454 KB
454 KB PNG
>>2762359
>>
File: May3.png (270 KB, 512x768)
270 KB
270 KB PNG
>>2762196
Thanks anon
>>
Is there a reliable prompt for patterns on skin? Skinsuit tag doesn't work as well with AnythingV3 as it does on NAI.
>>
>>2762420
You might have an easier time just inpainting them
>>
>>2762420
What exactly are you looking for? A bodysuit or tattoos?
Anything 3 and its derivatives typically work well with danbooru tags. In this context, "skinsuit" is a relatively obscure tag so I'm not surprised it doesn't do what you want.
https://danbooru.donmai.us/wiki_pages/skin_tight
>>
File: 01662-1081000775.png (1.75 MB, 976x1416)
1.75 MB
1.75 MB PNG
>>
>>2762427
>>2762431
Okay I'll try both suggestions, thanks.
>>
File: boob socks.jpg (38 KB, 480x295)
38 KB
38 KB JPG
>>2762003
newbie here, is it possible with one of these AI tools to remove "boob socks" and turn it into something normal like the middle example of this pic?
>>
File: naked-tabard.jpg (604 KB, 3072x1152)
604 KB
604 KB JPG
Anyone have tips for landscape oriented shots? Specifically reclining or lying down? The models seem to be overwhelmingly oriented toward portrait and most of my attempts with landscape end in nonsensical anatomy.
>>
>>2762528
Short answer: idk but that's an interesting idea.
It seems to me that the model would have to have a concept of this. I wonder if this can be trained on much like the "bad prompt" embed.
There is such a tag called "impossible shirt" that could be used in the negative prompt.
https://danbooru.donmai.us/wiki_pages/impossible_shirt?z=2
>>
>>
New to this, is it just me or does NAI leak make everything oversaturated?
>>
>>2762528
"shirt tucked into pleated skirt"
usually works for me
>>
File: PriestessAI_k.png (2.29 MB, 1024x1536)
2.29 MB
2.29 MB PNG
>>2762529
>>2762533
Please put Not!Priestess in ome of those.
>>
>>2762529
Just prompting "reclining" works for me
>>
>>2762441
I have generated so many tribal girls it's unreal. Technology is amazing.
>>
>>2762567
> oversaturated
If you mean too faded then use proper VAE
If you mean too vibrant, it's an anime model, of course it's vibrant, find some keywords to dull it down if you want.
>>
>>
>>2762357
catbox?
>>
>>
File: 01670-4220792061.png (1.57 MB, 1504x856)
1.57 MB
1.57 MB PNG
>>2762598
They really do hit different
>>
>>2762591
It's a naked tabard lora.
https://mega.nz/folder/OppA0QDI#w0_KOY0CyTpPtG6xopKwOw/folder/PppWkIjL
>>
Is there a retards guide for making a Lora of an art style? I understand the guides I'm looking at but they're all for characters. What tags do I add/remove if I want to specifically want just the art style. And for the images do multiple completely different characters help?
>>
File: dollgirl.jpg (579 KB, 3072x1152)
579 KB
579 KB JPG
>>
>>2762723
Usually you can leave tags alone for style LORAs. I'd add anything you don't want to neg against later, like monochrome, though the tagger usually picks up on that and adds it on its own. If the artist is only known for a handful of characters then you want to make sure those characters (or at least their defining characteristics) are properly tagged so you can change them later through prompting. The tagger is usually decent about this, but checking its work manually with a tag editor is never a bad idea.
>>
Do I need to prompt with controlnet to help it get the pose right?
>>
>>2762798
it helps, but the controlnet takes power over the prompt most of the time
>>
File: 321341411414.png (525 KB, 607x810)
525 KB
525 KB PNG
is there a lora or something for covered erect nipples .
like a nipple bulge/ puffy nipples visible through clothes , because no matter what tags i use its either full nude/breast out or 0 nipple silhouette .
being to do this shit for like a week now
>>
File: 00329-2817428396.png (307 KB, 512x512)
307 KB
307 KB PNG
>>
File: 00616-1491633245.png (326 KB, 512x512)
326 KB
326 KB PNG
>>
>>2762864
"covered nipples" should be the tag, but I've only seen it being used with emphasis like (covered nippples:1.1) or more see e.g. the nutmeg and g-anonmix exampled in the hdgrecipes rentry
>>
>>2762872
ive tried it a lot ( and "covered erect nipples" ) and the results seem to be very random . like half the time it would be with a little bulge the other time it would be like a sfw normal drawing ( 0 nipple erection)
>>
>>2762876
Have you tried "puffy nipples"? Maybe combine it with "covered nipples" as well if the nipples shows
>>
>>2762864
It's "covered nipples" for sure. Adding "nipples" in the negative does help reduce the amount of bare breasts but also seems to increase the amount of no nipple silhouette. After that, it's roll of the dice.
>>
File: controlnet tests.jpg (460 KB, 3840x1226)
460 KB
460 KB JPG
>>2762798
Yeah, openpose mode is designed for this.
>>
File: controlnet test2.jpg (371 KB, 2560x1817)
371 KB
371 KB JPG
Depth mode can also be helpful. You can mix multiple controlnets. To enable, go to settings -> controlnet -> Multi ControlNet: Max models amount and slide it up from the default of 1.
>>
psst hey anon...

Sexo...
>>
File: skinnydipping.png (1.86 MB, 1024x1920)
1.86 MB
1.86 MB PNG
skinny dipping on sunset
>>
I found a single Ram lora, but I didnt like the results kek
>>
>>2762996
Vae?
>>
>>2763040
braphog.vae
>>
File: 00007-4150175248.png (454 KB, 512x768)
454 KB
454 KB PNG
Has anyone ever managed to get "tan lines" to work? I'm not very experienced, but so far every model seems to interpet tan lines as some sort of white/light fabric.
>>
>>2763074
Probably put more emphasis on the tag than strictly necessary. Using AOM3-nutmegmix_40-pruned
>>
>>2763085
thanks m8, I'll give that a try
>>
File: poison.jpg (404 KB, 2366x1685)
404 KB
404 KB JPG
There is someone here that works with that img to image AI thing?

If there is, I wonder if you could put this through the AI
>>
How do I get SD to handle a name exactly as it's listed? I keep using Futaba Riho and getting Futaba Sakura...
>>
>>2763123
the are several ways :
use {Futaba Riho} instead Futaba Riho
use Futaba Riho + descriptive tags
use {Futaba Riho} + [Seishun Buta Yarou].
use a LoRa https://civitai.com/models/14533/futaba-rio-lora
>>
File: nep.png (1.34 MB, 960x1280)
1.34 MB
1.34 MB PNG
>>
>>2763122
OK. Now what?
>>
>>2762009
can i have a comparison like this >>2762009
with AbyssOrangeMix3 and bloodorangemix?
>>
File: file.png (1.25 MB, 960x1280)
1.25 MB
1.25 MB PNG
>>
Can you ask for AI requests on this thread or its just to post things you do?
>>
>>2763137
I'll direct attention to >>2762127 for those who missed it.
>>2763129
Is this what you're looking for?
>>2763124
You can also use underscores: futaba_riho. Also try putting futaba sakura in the negatives.
>>
>>2763148
exactly.thank you; it seems that ao3 is better can i know which version you use? i checked on civitai and there are many models (i mainly use inpaint) but i also like softnsfw (like image you posted) i also read that this model doesn't have its own vae should i use that of another model like ao2?
>>
>>
File: 00222-928210244-nsfw.png (1.08 MB, 1312x768)
1.08 MB
1.08 MB PNG
>>2763122
Your question isn't very clear, but I'm assuming you want to see if the AI can nude edit the photo or generate something new based on the poses.
Multi-subject images are a lot of work to get right. The more subjects the more difficult.
For instance, I didn't prompt for any particular characters here, but if did, their attributes would get mixed up. There are techniques that address this, such as Latent Couple.
https://www.youtube.com/watch?v=Sc9zK5Gj8no
>>
>>2763148
you people are amazing.
>>
>>2763150
They're simply listed as AOM3 and AOM3A1 here:
https://huggingface.co/WarriorMama777/OrangeMixs/tree/main/Models/AbyssOrangeMix3
I'm pretty sure the author intends you to use the orangemix vae for all of his mixes:
https://huggingface.co/WarriorMama777/OrangeMixs/tree/main/VAEs
>>
>>2763157
thank you again
>>
File: 00050-3448337247.0.jpg (473 KB, 2048x1024)
473 KB
473 KB JPG
>>
>>2763153
>Your question isn't very clear

I don't know, didn't expect anything on particular. Ive seen threads with img2img AI making and I just was curious what AI would do with that image I just saved.
I just feel curiosity for img2img AI stuff.
Thanks for your time in process it and thanks for the info. ITs really interesting.

As I say is just a subject I find interesting
>>
>>2763289
>>2763127 is also an img2img of your group pic - just a very different prompt, that just captures some of the light and dark shading in the original.
>>
>>2763122
what are you looking to get out of that pic?
>>
Wish I could do go ai art with ai waifu
>>
File: edit.png (3.79 MB, 1875x2500)
3.79 MB
3.79 MB PNG
>>
>>
>>
>>
>>2763223
catbox?
>>
>>2763334
Christ this stuff is getting impressive
>>
>>>/h/7243301

from me for you future anons, enjoy, ill be lurking for the next couple of days, if you got any requests of what to mix into this next then you can tell me
>>
>>2763329
What's going in here? Is this based on a source image? Care to explain the process?
>>
File: xyz_grid-0000-3435451575-.jpg (524 KB, 4000x1092)
524 KB
524 KB JPG
>>2763289
As >>2763291 points out, img2img results can vary wildly depending on what you want. Basically what it does is add random noise to the input image and use that as its seed for a txt2img prompt. Depending on how much influence you give to the input, you can see why asking someone to just use the image is a bit like asking someone to write a sentence, any sentence at all.
>>
File deleted.
>>2763400
use saucenao next time before post
>answer
Is based on picrel
>>
>>2763395
Greetings, anon. I'll be sure to add your mixes in when doing batches.
I haven't been following what's going on in /h/. What are you aiming for with these mixes? General capability, a particular style, just having fun?
>>
>>2763417
just had fun with mixing and wanted to learn how to do things, i find it interesting in your comparison photo how it doesnt break the face too much even when trying to do those squinted anime eyes on the top row, its supposed to be semi-anime so it can do anime expressions but clearly not to that extent that the more 2D models can
>>
OK maybe this one's not the best quality but it's fun.
>>
File: anna.png (780 KB, 768x1152)
780 KB
780 KB PNG
https://mega.nz/folder/ix0XzJ7T#BpVEvZKXLEVoiU87FhehKA

enjoy
>>
>>2762127
If its not much impossing and someone wants I could dig into some Garnet Til from FFIX (long hair) images
>>
>>2763400
hereĀ“s the process:
>paint over the original like a 5 year old.
>put it into inpaint and use interrogate to get a couple tags. remove the ones that dont fit and maybe add some.
>inpaint it and generate couple of times with varying degrees of denoising
>>
>>2763387
thats an edit i just decensored the pussy. look up unfairr, he does great stuff.
>>
>>2763501
you dosent need this for mostly images

although the process I use does not work if hands/objects/are close or interfere as in this case
>>
Learning the correct prompts to make a short stack.
>>
>>2763567
use a LoRA and/or embedding. also adding "tall" to the negative prompts works for me usually
>>
>>2763567
There's a shortstack lora out there.
https://mega.nz/folder/IPM2ERrL#a6Zti0k0G7FZmQz2AfCgrg/folder/8SdkGRIQ
The same anon has goblin and incase loras that also do well in the shortstack category.
>>
>>2763574
I really gotta catch up on the new stuff. I kinda just setup NovelAI and have been having enough fun with that
>>
>>2763152
Super cute need more of them
>>
I tried the lora training guide but the scripts are oriented both for Windows + Nvidia. I'm on Linux + AMD so I didn't feel like troubleshooting when python barfed. Guess I'll stick to being a user for now.
>>
File: 00054-4152573272.png (1.57 MB, 1024x1536)
1.57 MB
1.57 MB PNG
>>
File: toph-sketch.jpg (619 KB, 3072x1152)
619 KB
619 KB JPG
>>
File: samus-sketch.jpg (766 KB, 3072x1152)
766 KB
766 KB JPG
>>
File: download (1).png (334 KB, 384x704)
334 KB
334 KB PNG
>>
File: birthdaysuit.jpg (2.08 MB, 3237x4096)
2.08 MB
2.08 MB JPG
Just learned about openpose editor, posing would be doable without being constrained with text prompt.
>>
>>2763751
What's up with ESLs and really poorly trying to caption porn with the most broken English known to man? If you don't know English just write the text in your native language and let someone else translate it if they care.
>>
File: 00015-2624328722.png (2.12 MB, 1536x1024)
2.12 MB
2.12 MB PNG
did anybody have any success with prompting decent ENF content with multiple characters being involved?
it seems almost impossible
>>
>>
>>2763770
Needs a vae
>>
>>2763751
I'm gonna have to look into openpose. I'm only just now getting the hang of controlnet. This stuff is changing faster than I can keep up with.
>>
>>
Why is it so hard to generate facials?
Or rather, anyone have any better ideas than spamming ((((facial))))? I understand faces are complex, and cum on them probably looks like random noise to the model, but any ideas?
>>
>>
>>
>>2763827
Strong emphasis doesn't really work for me. If you have to strongly emphasize it, the model is very weak at it to begin with. What model are you using? Try something more hardcore trained and use a wider variety of tags that get you there. For instance, sweat might accidentally get you there since it might fight between drawing sweat and cum.
>>
Any tips for generating dakimakura art? I want to print some if possible.
>>
File: bocchi-upskirt.jpg (488 KB, 3072x1152)
488 KB
488 KB JPG
>>
Is there a way to run this remotely and privately? My laptop can't handle it.
>>
>>2763889
Yes, but you're going to need to buy some compute and be able to go through the installation.
>>
>>2763839
>>2763845
Noob here, trying to get to grips with sd. What model was used for these?
>>
>>2763889
https://rentry.org/sdg-link#run-cloud-hosted-instance
OP here. After going over to /g/ I realized that this a better getting started resouce than sd-mashup. I'll be adding this to future OPs.
>>
Anyone have any good recommendations for generating a good Android 21?
None of the Lora's I have or prompts I seem to use want to make an accurate Android 21.
I use the anything-v4.5 checkpoint
>>
File: Previews.png (2.61 MB, 1735x856)
2.61 MB
2.61 MB PNG
new models, test them out, looking forward to seeing the results, comparisons specifically

>>>/h/7248513
>>
>>2764180
This is UltraStabbedHeaven - I'm impressed that on the same prompt and seed that just gave 1girl on AOM3/nutmeg40 it understood aircraft.
>>
>>2764188
>prince of turbojet
>English planes
ill add "mixes can produce airplanes" as a feature
>>
>>2764191
Contrast AOM3/nutmegmix40 with the same prompt and see, though
>>
>>2764165
She doesn't seem to be well known by any of the models I tried. This is the best I could get out of 7th Anime 3C.
https://litter.catbox.moe/za4ti8.png

The prompt uses this lora >>2762236 https://files.catbox.moe/5g988b.safetensors
>>
>>2764234
D'oh I quoted the wrong lora. >>2761922
https://files.catbox.moe/v11kre.safetensors
Pic related is same exact prompt with DPM++ 2M Karras instead of Euler a.
>>
>>
>>
>>
>>2764180
Anon, 9 models is a lot to evaluate at once. I personally keep about 6 in my rotation when doing batches.
>>
>>2764352
the fruit ones are based on grapefruit
in your comparison youre using aom and anything, as AOM is a mixture of anything and basilmix that is fine

my older mixes using mid preset were based on AOM, it is probably best you compare the AOM based Stabbed Abyss / Super Stabbed Abyss / Ultra Stabbed Abyss in that case

im not expecting you comparisons for all 9 models, any comparison is welcome
>>
>>2764352
thanks also, it looks good so far
>>
>>2764352
any new 2d Nai model based?
>>
>>2764356
thats a great idea anon! i should create a NAI based version, ill make them right now and upload
>>
File: grid-0260.jpg (427 KB, 2304x2304)
427 KB
427 KB JPG
>>2764180
Some no positive prompt tests on UltraStabbedFruit. I've run more than just what's shown but it's very representative of the other results. I guess the model is good for nude blonde girls with cat ears.
>>
>>2764365
ive forgot which mix does it but ive read of somebody else's mix that seems to have a tendency to create catgirls when not prompted properly

it seems catgirls are universal, just like elon musk wanted
>>
>>2764367
made the models but cant highres my previews so they are the same settings as before due to current auto1111 having issues with memory, i used to be able to 1080p but now i cant even do that with yesterday's updates

what i can say though is that NAI is a very good base, ill wait until auto1111 fixes its issues before showing these off
>>
File: grid-0262.jpg (694 KB, 2304x2304)
694 KB
694 KB JPG
UltraStabbedFruit + Lum lora. I really like how the bottom left one came out.
>>
USH generated a rather bored looking demon
>>
>>
>>
File: 00159-4002958587.png (953 KB, 640x960)
953 KB
953 KB PNG
>>
File: 00161-3932031149.png (810 KB, 640x960)
810 KB
810 KB PNG
>>
File: 00177-3429245890.png (1.2 MB, 1200x750)
1.2 MB
1.2 MB PNG
>>
File: 00074-3904605469.0.png (487 KB, 480x768)
487 KB
487 KB PNG
>>
File: 00075-3904605469.0.png (490 KB, 480x768)
490 KB
490 KB PNG
>>2764493
>>
File: tmpqq4olqq6.png (1.28 MB, 1024x1024)
1.28 MB
1.28 MB PNG
>>
File: 00733-1158228001.png (259 KB, 512x768)
259 KB
259 KB PNG
>>
>>2764357
will you release it on civitai?
>>
>>2764573
nah fuck those retards, i attempted uploading yesterday but then got cucked for forgetting to set copyright permissions according to what aom3 and grapefruit, instead i had them set so people could do whatever they want with them, meanwhile the AOM3 faggot specifically wants his shit credited and not used in X or Y (fuck those retards mixes should be open source no matter what anyone says)

so no im not releasing them there, just a waste of time reuploading again, people who use civitai are subhumans anyway

fun fact: for the time my mixes were on civitai which was about 30 mins i had already gotten 50 downloads and people were saying they were good as fuck
>>
File: 00021.png (3.8 MB, 1280x1920)
3.8 MB
3.8 MB PNG
Basic question by a fairly new SD user: When should (tag) ((tag)) (((tag))) etc. be used and when (tag:1.2) and so on?
I've also seen things like
tag a, tag b, (tag c, tag d, tag e:1.2), tag f..
>>
>>2764680
multiple brackets are old style
explicit multiple is a new style (()) -> :1.21 ((())) -> :1.331
Bracketing together tries to suggest that things are associated with one another, though I don't know how much it's just placebo.
>>
>>2764356
Anything4.5 (in name only) has less of the hyper real skin shading, at least enough to counteract with [3d] in the negative prompt
>>
we need a conan anime
>>
File: 00020.jpg (300 KB, 1280x1920)
300 KB
300 KB JPG
>>2764685
Thanks, that helps.
>>
>>2764687
But one where he fights Wrath-anon, not the shitty one with the kids
>>
>>2764686
just tried it I gave it a quick try (so I did not fix mistakes) and it works thanks anon
>>
>>2764655
What about a torrent once you've settled on your best handful of models? Torrenting is my preferred method of getting a model and I seed them indefinitely. I don't even have a civitai account.
>>
>>2763490
I couldn't get this to work on any of the models I have. She's just not strongly represented. Considered making a lora but I haven't figured out how to make loras on my setup yet.
>>
>>
File: 00016.jpg (1.63 MB, 2304x3584)
1.63 MB
1.63 MB JPG
>>
File: 00429-2676115812.png (1.44 MB, 888x1328)
1.44 MB
1.44 MB PNG
>>
>>2762043
Oh, that's the closest I've seen to the fingers actually going inside.
>>
Is there a lora or something for sheets of cloth covering a naked body? I can't recreate this without img2img.
>>
File: 00038-881348272.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
I'm using Latent (nearest-exact), but I get these weird blurred edges...
This seems to be the best upscaler to not lose image details, and to remove it I need to do a second pass with img2img and that changes some parts of the picture (and takes time), even if it removes these...

Any fixes or similar alternatives? No other upscaler does this.
>>
>>2764859
I'm not an expert on upscalers so I don't know the difference between the various upscalers, (I mainly use inpaint and use lanczos when I have a very small image) but have you tried doing a 512-512 >extra> and then doing an upscale?
>>
>>2764859
model? denoise strength?
>>
File: file.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
>>2764862
I prefer upscaling while making the picture to give it more detail

>>2764866
>denoise strength?
shit, that was it
I was using a really low value (0.45) and 0.7 seems to do the job for now

enjoy the pic
>>
>>2764867
0.54-0.63 seems to be the sweet spot for me. I've used 0.57 forever now. Sometimes I use remacri if the latent is fucking up a particular image.
>>
>>2762629
sorry for late response didn't see your reply
https://files.catbox.moe/kicg5x.png
>>
File: 01709-3774682342.png (1.93 MB, 1016x1400)
1.93 MB
1.93 MB PNG
>>2763611
I had to install W10 on another drive to get lora training to work too
I also had to make my pagefile like 80GB kek
>>
>>2764992
I never would have thought that Windows + Nvidia was the best option for AI training, but here we are.
>>
File: 900.png (887 KB, 768x1024)
887 KB
887 KB PNG
>>2762528

The AI's been able to generate untucked blouses, etc., for a few months now; you just need to specify loose or baggy clothing, and avoid tags like "skindentation."
>>
>>2764705
thanks for trying

>>2764729
these anime style AI generated images I've been seing lately are really impressive, I would never guessed those are forged.

Up until now all ai had that distinctive style coloring that gave them away but now Im seing plenty of more anime like pictures that make really hard to tell
>>
>>2764868
yeah, I've been doing 0.575 for the past day...
reporting results:

>latent (nearest-exact)
>using 0.575 completely removes the blurry edges and such (while trying to not change the
original picture too much)
>degrades quality a fair amount
>also changes some areas of the original picture
>original picture is the best output, but needs to be upscaled for actual best results
>trying out other upscalers give me a subpar result, and/or lack of detail (and quality overall)
>doing a second pass with img2img also removes the blurry edges AND also changes the picture a bit (hence loss of quality)

I honestly don't know what to do, it's very clear latent (nearest-exact is the best, but having to deal with the blurry shit is a pain, and the only workaround is forcing the denoise check and pray the rng is good enough to get a good output, and we're already dealing with rng for the original pic anyway...
>>
>>2764493
>>2764494
who's this qt?
>>
File: 05232.png (864 KB, 720x960)
864 KB
864 KB PNG
>>
>>2763755
Many people learn english through porn.
>>
>>2765210
>Up until now all ai had that distinctive style coloring that gave them away
I agree. I think for the most part, people favor the glossy and realistic models because they have really nice details, but they all have that kinda generic AI style. Now with specialized models and style loras you can really dial in more flat and hand drawn feels.
That model is specifically 7th Layer Anime.
>>
File: mina-sketch.jpg (1.34 MB, 3072x2304)
1.34 MB
1.34 MB JPG
Not sure how I feel about these but they're interesting nonetheless.
>>
File: 00170-3894105666.png (849 KB, 768x768)
849 KB
849 KB PNG
>>2765341
There's a vinyl sticker lora on civitai if you want to test that out.
>>
File: 1676471420962475.png (1.7 MB, 1024x1280)
1.7 MB
1.7 MB PNG
Needs more beautiful backgrounds to go with the women
>>
>>2765270
why are you trying to make her 3d bro
>>
File: anna_yamada_22.png (830 KB, 768x1152)
830 KB
830 KB PNG
>>
File: 00074-3137621050.png (1.36 MB, 1000x1000)
1.36 MB
1.36 MB PNG
why are the nipples dark?
>>
>>2765503
They're slightly tanned from wearing nipple hole shirts and swimsuits all the time?
Try a pink nipples tag maybe?
>>
>>2765503
I might be able to guess if I knew the model and prompt. The short general answer is that the prompt is probably influencing the color palette.
>>
>>2765264
Well, I'll be lying if I call her Sakie Sato from Demi-chan wa Kataritai because that doesn't look like her, but I did insert her physical features/clothes + dark skin with a Pepe artstyle lora

https://mega.nz/folder/TZElCCIL#OEutCsJPD0G5lXPXLGgfcQ/folder/XVkjBRhS
>>
File: 00103-1418417166.png (776 KB, 768x768)
776 KB
776 KB PNG
Recently started to make ai art and download anything v3. Am happy with it, but I how do I go from the picture I have attached which I made through anything v3 and something like this?
>>2762011
or this?
>>2762009
>>
>>2765602

And this
>>2765011
>>
>>2765602
Have you considered NOT using Anything V3 I'm almost 100% sure >>2762011 use a different model, I don't know what, but definitely not Anything V3
Which model should you try you might ask; well that would depends on what you like (flat 2D, 2.D with prominent nose, etc.)
You'll can get a lot of different recommendations from different anons if you tell them what you'd like to see

As for the MHA character, it must have used a lora, check the gayshit lora list in the OP and ctrl+f her name
>>
>>2765602
Mina Ashido is a lora and I typically use it at 0.7 strength. Pic related is in fact done with Anything v3. https://litter.catbox.moe/7v59s7.png
>>2765011 was done with AbyssOrangeMix3 A1 and has two loras: slingshot bikini and shortstack. It can be done with Anything v3. Here is the same prompt with Anything 4.5 (note that 4.5 is an opportunistic name to grab attention, it's not the true successor to Anything 3) https://litter.catbox.moe/rzivif.png
For the loras, just search the lora list in the OP.
>>
And to expand a little on what to do to improve your prompt, here's a Tsuyu made from Anything v3 with no extra loras or tricks or anything. My guess is that the big difference is the negative prompts. Download the catbox version and load it up in your PNG info tab to see the prompt metadata.
https://litter.catbox.moe/wg4ar4.png
>>
Pro tip: Edit "txt2img/Negative prompt/value" in ui-config.json to include your favorite baseline negative prompt. That way, it's there as a starting point every time you load up the webpage.
>>
File: 00418-3898327839.png (1.89 MB, 1294x1294)
1.89 MB
1.89 MB PNG
chu
>>
File: 00001.png (1.18 MB, 1024x1536)
1.18 MB
1.18 MB PNG
cute butt
>>
File: 00001-440477956.png (2.28 MB, 1024x2048)
2.28 MB
2.28 MB PNG
>>
How feasible would it be to have a database of only hand pics, then train an AI to look for hand-like parts of a picture generated by less handy general AI and replace with one from the database by inpainting?
>>
>>2765739
This has basically been done with controlnet. I haven't personally messed with it yet.
https://www.youtube.com/watch?v=Oqf-93TpykI
>>
>>2765741
So it's just a matter of plugging that in the processing loop?
>>
File: 00022-3224047419.jpg (3.65 MB, 2880x2880)
3.65 MB
3.65 MB JPG
>>
>>2765746
Like I said, I haven't used that specific library, but controlnet in general works by influencing generation using other existing images. It has broad capability ranging from referencing just the pose to tracing the reference image. I think with the library, you choose hands as reference images.
>>
File: 00081.png (966 KB, 720x960)
966 KB
966 KB PNG
>>2765270
>>
File: 00003-3377505934.png (1.97 MB, 1024x1536)
1.97 MB
1.97 MB PNG
Just got started, combining gens with inpainting and photoshop. Already really happy with results
>>
>>2765815
Koharu 80s?
>>
File: 00019-4000016418.png (680 KB, 512x1024)
680 KB
680 KB PNG
The Anything-V4.5 model doesn't seem to understand lewder prompts like "female masturbation" or "hands touching panties" for some reasons
>>
Anyway ai can do ranma like art?
>>
>>2765838
yeah 80s and koharu lora both on civitai
>>
>>2766000
You mean like ranma 1/2? AOM3-A1 ain't half bad at it.
>>
>>2762011
>>2762047
>>2762050
>>2762213
How do you get lips to look that good? I've been trying to give my images nice full lips like that to pretty mixed results. Is it a specific Lora you're using?
>>
>>2765597
I see, thanks!
>>
File: 00113.png (465 KB, 512x768)
465 KB
465 KB PNG
>mixing so many different loras i dont even know what board to post to
>>
File: file.png (2.42 MB, 1024x1536)
2.42 MB
2.42 MB PNG
>>
>>2766069
>me-ow!
>>
File: catbox_8k8k5k.png (1.82 MB, 1024x1536)
1.82 MB
1.82 MB PNG
>>2766071
Meow indeed
>>
>>2762003
Trying to download one of the lora's, but it get stucks on 14.48/14.5. Anyone knows what can fix this?
>>
File: 1650091312015.png (1.31 MB, 1024x1536)
1.31 MB
1.31 MB PNG
>>
File: 00011-848598501.png (1.71 MB, 1024x1536)
1.71 MB
1.71 MB PNG
>>
>>
File: 1670029761995750.png (2.62 MB, 1380x4009)
2.62 MB
2.62 MB PNG
>>2766021
Yeah and I wish I could used it for this (pic related) transfer but I'm on mobile
>>
>>
>>2766082
Steven Universe? You lost me, anon. I have no idea what you're talking about.
>>
>>2766098
I've been trying with some mobile apps to do that image of lapis in the art style of ranma but it always comes off horrible
>>
is there an easy way to get a consistent character across multiple image gens? for example generating a character then doing more generations with that character wearing different outfits, hairstyles, or doing different poses but still having the same face and body size/shape
>>
>>2766103
The artist that does Ranma is Rumiko Takahashi (same as Lum / Urusei Yatsura). Perhaps you will get better results if you try "by" or "style of" Rumiko Takahashi.
>>
>>2766109
Assuming you're talking about a new character from scratch, not an existing IP. If you stick to the same model, sampling method, and characteristics (hair color, eye color, etc), you can more or less get someone that looks the same with some luck.
If you need to take it a step further, a lora would probably be best. You'd have to have a handful of images as a starting point, then use the lora to gen more images. It would probably take multiple iterations, but you can gen more images and feed back the best results. I'm pretty sure anons in /g/ and /aco/ have done this before.
>>
>>2766153
Yeah that's what I ment
>>
>>
Does anyone know a good model to generate pictures of ben 10 in the original art style?
>>
>>2766199
You're probably best off training a LoRA on existing official material
>>
>>2766199
Pretty sure /aco/ has done this already. Might want to ask around there.
>>
>>
>>
File: 00004-383864195.png (596 KB, 632x800)
596 KB
596 KB PNG
>>
File: 00005-4164894110.png (621 KB, 656x800)
621 KB
621 KB PNG
>>
File: 00009-3998655598.png (655 KB, 800x656)
655 KB
655 KB PNG
>>
File: 00011-3998655598.png (683 KB, 656x800)
683 KB
683 KB PNG
>>
File: 00014-759499221.png (708 KB, 656x800)
708 KB
708 KB PNG
>>
File: 00015-2280635170.png (793 KB, 704x800)
793 KB
793 KB PNG
>>
File: 00016-1403856178.png (746 KB, 704x800)
746 KB
746 KB PNG
>>
File: 00017-173666988.png (765 KB, 704x800)
765 KB
765 KB PNG
>>
File: 00019-325768027.png (761 KB, 704x800)
761 KB
761 KB PNG
>>
File: 00026-295775781.png (774 KB, 704x800)
774 KB
774 KB PNG
>>
Only when you decide to blow op a pic, do you start to see its faults. Still love this one but now it makes me want to pick up Photoshop.
>>
Also, any tips on getting really good breast that are basically falling out of clothes? This came out really good, but I want to be more consistent
>>
>>2766046
I think it's the akira lora
Pretty much all her artwork I used to gen it was heavy on dick sucking lips
That + "lips" in the prompt usually does it
And for example this >>2764992 was like (lips:0.5) I think
>>
NO ONE LIKES THIS TRASH. notice how there's no discussion here, just you retards posting shit no one wants to see.
>>
>>2766300
Catbox please?



[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.