>>1864

>Waiting somewhat impatiently for someone to create a de-breasting LoRa. It's been my fetish for some time.

That makes two of us!

>So far I've only seen one debreasting posted here. Pretty nice one too.

I must have missed that one. Would you link that pic, please?

It appears that someone on pixiv may have something, but they haven't uploaded anything for some reason: https://www.pixiv.net/en/artworks/119277886

There are some more such supposed lora posts on that profile but I couldn't find any links for those either and the poster doesn't respond to comments.

Also I found this on pixai, but it didn't work for me. https://pixai.art/model/1719210777597320429

(...hope I didn't just post the link for your own attempt back at you with this)

I always wanted to do a debreasting - lora, too, once I was done with the pussy stabbing thing that Entropy now finished (>>1800).

So I have a ton of potential pics for training. But after seeing how it and other Loras behave I'm beginning to suspect that it may be necessary to seperately train multiple debreasting related concepts in order to achieve good results; one lora for a single amputated breast, one lora for both breasts amputated, one lora for cutting a single breast, one lora for dual removal and so on. The severed breast (laying around, falling off, beeing held) might also need to be trained as seperate concepts. But that is just me speculating.

But yeah, I agree, this totally needs to be a thing!

...oh and here, hope you like this old redo of a classic carl pic I made before dolcettish was taken down. :)

I made a dbereasting LoRA just last week.

Still running test on it. will post soon

>>1864
for example that pic was in the dataset and this is how it looks when i try to recreate it.

>>1868

That's awesome! :o

These pics look very promising!

>>1865

The one debreasting I saw was in the pussy stabbing thread. The comic with the cute little lady getting chased down by Spanish soldier, stabbed in the cunt and throat cut, then they cut off one tit just for giggles.

https://boards.guro.cx/ai/src/1717993189312.jpg

>>1883

Oh! I made that one. xD ...I thought someone else had also made a debreasting pic that I just hadn't seen. ^^ ...happy to hear someone likes it. :D

TAM TAM TAM

here you go guys

https://www.mediafire.com/file/yipak5r4qlunr9q/debreast_LoRAs.zip/file


It was done by me with the help of Entropy

https://www.pixiv.net/en/users/32339993

please post your results

>>1909

Kaboom! Yay, you did it! :D

Testing now, so far this is what I've got...

>>1909
Nice! Are there rules on how to use it? I mean, does it match with a specific resolution, only do anime, things like that?

>>1912
For resolutions just as a rule of thumb you want be using multiples of 64. For SDXL, don't go below 512 for any one dimension. I had a look through the data set and anime style images seemed to be the most common. But it should work with any style, while I haven't tested it I've made Lora's with purely anime images and it worked with other styles just fine.

>>1913
>>1909

Also I've been really struggling to get "debreast_fly" to work. Do you have any recommendations for it? It just never appears for me.

Thanks for the amazing Lora!

>>1914
Unrelated to the discussion but just a funny gen I got. Looks like boob monster

that is the true beauty of AI

PONY+PS such an amazing effect.Thanks for making this lora.It really takes a lot of time to make a big picture.
PONY+PS 令人惊叹的效果,感谢制作这个lora，作一张大图真的好花时间。

>>1909

Thank you for the Loras! Bit tricky to get them to behave, but when they do they deliver pure gold! :D

So far I'm having the best results with the debreast_mutilated version...

(the only inpaint/manip done on this example were the additions of the breast in the pan, the handle of the pan and some more blood on the knife)

...this one was the very first prompt with the debreast_fly Lora on a new model I've found yesterday. Same prompt as the last image I posted, only made a HiRes pass with Adetailer for the face and then upscaled once more...

...So far since I got pony running this is only the 4th model I've experimented with so I know I'm still fairly noob at it. But at least with the debreasting Loras, the pussystabbing Lora and the quad amputee loras I have gotten far better and more consistent results with it than with PonyXL, Nub mix and autism-mix.

So I really have to recommend you guys check out that model! (Unless of course you know of even better models, in which case, share your knowledge, please! :D)

Look for "icatcher" on civit. there are 4 versions. These 3 example pics where made with the cartoon version. Semi-realistic also works really well. I haven't gotten around to test the anime version yet...

(Grid pic uploaded just to showcase how consistently the concept gets represented. With the other models I maybe had 1 or 2 pics out of 10 where there would at least be one flyer)

Will post some more examples here and in my thread soon.

>>1913
>>1914

These are nice! Who is she?

>>1915

Yay, nightmare fuel! xD

>>1919

Also very nice! What do you mean by PONY+PS? Pony and photoshop?

>>1922

>Same prompt as the last image I posted

Oops... I meant same prompt as >>1911

This only took some minor inpainting.

1st) a remaining nipple was removed and both stumps made a little nicer. (Check meta for crude but nicely working inpaint prompt blockweight (only steps) mix of debreast fly/debreast_mutilated. Put denoise higher though, meta is from last tweaking pass with unchanged prompt)

2nd) one surplus flying breast was removed and 3d) crude yellow fatty tissue rings sketched in and then finalised with img2img and upscale

I can not thank you enough, anon from >>1909! Simply amazing! :)

...Paradise!

...That's what this Lora is!

.aaand yes, the anime version works well with the lora, too! (although still the best results came from the cartoon version imho.)

Just txt2txt with heavy adetailer use for the two hands. Yeah I know, they still suck. xD ...I'm just too tired to finish this pic now, just wanted to showcase the anime version. Maybe tomorrow...

>>1928
those are amazing. much better that wha I was bale to get, and i was the LoRA creator.

I really need to stop generating tons of pics and focus to have a few perfect ones.


I'm really happy you are making pics with it.

If you create enough good quality ones I retrain the LoRA to a new better version.

I would like to see a breast grilling lora

>>1930
Can you do that to Nami of One Piece?

>>1869
Oh you did, my bad. Maybe do this to Uraraka?

>>1932
you can already do this with no need for additional LoRAs

this what I did a long time ago

>>1931

>those are amazing. much better that wha I was bale to get, and i was the LoRA creator.

Thank you! Really happy you like them! :D

These results are in no small part due to these models I recommended
yesterday!
I'm particularly amazed at how aggressive it let's you prompt weightwise. Maybe I've just missed something with the other pony models I've tried but all three of them gave me garbled blue static as soon as I looked at either prompt- or Lora- weight the wrong way. None of them took any lora I tried at more than :1.1 and most triggerwords wouldn't allow me to go higher than :1.3. I even got these blue glitchfests when I just moved single unrelated triggerwords elsewhere in the prompt from outside some adjacent brackets to within while not even touching the weight.
With this model however I've gone as high as :1.8 for the Lora and :1.8 for prompt as well in one case before getting the blue nonsense again.

Pic related... I swear there's a debreasting in there somewhere... >.<

>>1931

>I'm really happy you are making pics with it.

>If you create enough good quality ones I retrain the LoRA to a new better version.

Awesome! I think it would be best if you could tell me what kind of pics you'd need the most for the dataset Hit me up on pixiv, maybe we can make some kind of a to-do-list or something. (If we haven't had contact on pixiv already you can find my profile in the readme from this Lora: >>1800)

>>1934

I'm not the guy who did Nami, that was the creator of the Lora, I believe.

But sure why not?... One Uaraka coming right up! ...You want some fries with that? xD

Gotta catch 'em all! Oh, wait... wrong franchise...

>>1941
Looks tasty yum! Thank you. Could I possibly get an order of frog legs and breasts meat from Tsuyu.

...I wanted to see If pony knows who 7of9 is since there is no lora for her (only 2 for SD1.5). Aparently it doesn't.
Maybe she could pass as a Borg in whatever poses as star trek nowadays, I don't know.

So let's just call her 0of2...

Debreast_fly really likes to put a scratch and blood on most gens cheeks for me. The first Uraraka pic had one too that I removed. But with this pic it gave me the opportunity to cheaply get at least a tiny little bit of cyborgness going on with her. Just desaturated the scratch down to zero and added some colorspecs as grebly blinky light details or whatever. Also added just some colorspecs and techy lines to her pupils by hand and then left it as is.

That Bot on the left was placed there so conveniently, that I only had to add that tractorbeam and the blue lens and inpaint at super low denoise.

>>1943 Frog legs? oO ..I'll have to take a look at the character. I haven't watched hero academia. That's also why I only caught the fact that basicly all details from Uraraka's outfit where missing from that first pic when I thought I was almost done with it. Had to add and inpaint everything that is pink so she wouldn't just wear a blue bodysuit. I know her belt thingy is still missing but I believe if I had tried to add that I would have messed up everything on the tray. (Like I said I haven't watched the show, just know her character from other guro art, so I don't know if it's even canonically possible to detach that stuff from her outfit. If not, my bad, sry!)

>>1935

Not that requester but could you elaborate on that a little, please? Just "breasts on grill" or how do you achieve these?

Can I second the request for Nami having her massive tit's cut off?

>>1944

I suck at this so bad. Haven't been able to achieve anything with this LoRa. It just won't remove tits from the subject. I don't know what I'm doing wrong.

>>1935

How about breast burgers?

>>1989 I wish I could help you. D: (Assuming you're the Lora creator), you're working with comfy correct? I've taken a closer look at your image's meta in the Lora folder. I don't see any brackets and weights for individual prompt words there. Is that not possible with comfy? Or does it just work differently? I've tried your unaltered prompts from some of your pics but I too don't get much good results if I don't increase the weight for the Lora itself to at least :1.2 and then emphazise the main triggerword with weights above :1.3. Also further emphazising "wardrobe malfunction, guro and blood" a little helps.

<=== this raw txt2img pic for example needed both the Lora set to :1.5 as well as emphasis for (debreast_fly:1.5) to work. Other than that I only used hires x2 @0.24 denoise and adetailer on faces. I only let the prompt run for 3 gens once the weight values were high enough. The other two gens were as good but since here the stumps appeared on the phone screen all by themselves I had to take this one of course... (metadata included in pic in case it helps. >>1922 and >>1930 also have meta)

...

...and just a little tweaking later...

Debreast mutilated works good at lower weights but still, it too needed both emphazising the triggerword and Lora weight at :1.2 to give me something nice with the chefs prompt from the previews in the Lora folder. I also had to split the prompt for the female and male char and add "girl chef with moustache" in the negative prompt, because that's exactly what I got at first... or men debreasting themselves... >.<

But now its giving me these non stop. There are sooo many good ones, I can't decide which one to pick so choose one from the grid for me to refine if you like. :)

All the preview pics in the Lora folder where made with Pony_V6, if I'm not mistaken. Do you only use PonyV6 for your gens? I didn't have any good gens with Pony at all (though I got frustrated with it quickly and have only tried it out briefly). But as per statement by the creator of Pony all guro pics had been removed from training. Just speculating here but I read elsewhere that earlier Pony models had a ton of guro (albeit mostly furry). So maybe the models that work better have just had older versions of pony incorporated or where further finetuned on datasets containing guro.(?) But yeah, I definitely recommend you try something that's evolved far from base pony...

(P.S. Did the Dataset contain like 10 copies of that classic carl pic with the chef asking her if she wants a piece of herself? Because the Lora is really good at this scenario and the preview pics you made all look a little like it, in my eyes. :D )

Also I found that some Loras have the char's tits hardcoded in really really strong and mess with the debreast lora.

For instance, (and @>>1933 and @>>1946) one of the three nami loras I've tried so far is giving me a headache. Of course it has to be the one that is best at depicting the original style of the show. But I think I've got some nice ones in the making for you. I've been busy irl for a few days and will be some more. So it may take a few more days day or so, but I already have at least two Nami pics that I want to refine. (and so much other stuff)

But for now 1st I've got bratty littly Liru loosing her puppies for you to enjoy. I wanted to try making things a little bloodier this time. I'm not sure about this one, because I've stared so long at it, while waiting for the inpaints to finish, that I've lost the ability to judge if it's ok or not. So some honest opinions would be welcome... :)

<=== >>1943 Thanks buddy... I looked up fried froglegs for reference... -_- ...almost threw up over my keyboard... xD

(I knew they're a thing, just never saw photos of them before... I know you most likely wanted HER froglegs on the menu instead, but that would have been even more tedious work than it already was. Also I know her goggles aren't correct but since the pic isn't even close to the shows artstyle anyways I hope you don't mind.)

Oh, also @1992; jepp, breastburgers are a thing now. (more to come soon™) Apparently debreast_fly knows what Breastburgers are. Just prompting for (Breastburgers on table) with high weight (in this case :1.6) gives good result in around half to 2/3 of gens. On these 2 examples of burgs here I only refined the nipple stump - area. other than that they just received one low denoise img2img pass to enhance them just a little.

Still this took quite some time to make. Here too the Tsuyu Lora had to be weighted down in order to get workeable results, so that it also lost much of her likeness which I then had to restore via manip/inpaint.

So I hope you guys enjoy your romantic Breastburger - Dinner with Tsuyu with a sidedish of nipples and frogleg salad

>>2029
>Also I found that some Loras have the char's tits hardcoded in really really strong and mess with the debreast lora.

It could have to do with size/shape also. I've noticed sagging breasts really want to "weld back" into place, for example.

Have you tried other samplers like Karras over Euler A?
Karras feeds noise, which can result in more unstable anatomy when your prompt is too loaded, but it might work better for this.

I'll give it a go when I'm home

>>1946
can I third the request for Nami having her giant breasts removed?

Thanks for the tutorial.

I;m not the guy that requested it but I will try to produce soemthing nice with it.

>>1933
>>1946
>>2061
Not that guy but here you go

>>2063
Guys please post more high quality debreasting poctures here

Especially when there are tits flying


I willntrain a better quality debreasting lora once we have enough

>>2112

Frantically working on it! I'm finalising a few right now, will most likely upload in a few hours.

Would it be possible for you to share the datasets you've used? That way I think we could better see what kind of pics we need to make to help you improve the Lora.

Also I have collected a ton of debreasting pics over the years. I don't want to spam you with duplicates, so I'd love to compare and then send you what I've got (of course only the stuff that matches the quality of your trainingdata)

For now this is just the girl from >>1862 but without the sword and cum on her face (in case that works better for training)

>>2063 nice! :)

>>1933
>>1946
>>2061

Got 2 Nami pics for you. With this fist one again I tried to do something more bloody and kind of dynamic. Don't know if I succeeded, same problem as with the Liru pic before. I'mo longer abled to tell if it's ok or utter shit. If it's the latter (as I kinda suspect), I'm sorry!

This second one I'm much more happy with. I know, they don't fly here, sry. But from what little I saw of One Piece I'd say I can believe Nami would do this... (PSA: if a girl has crazy eyes like hers, run away and don't turn around!)

(The "?!" speechbubble in the first pic appeared as is by the AI's doing. Don't know if that comes from Pony in general or if the Lora was trained on Mangapics but I thought it fits the style so I lazily copied that sceme for this one here.)

I got at least one more of her in a completely different style that I'll work on later, so stay tuned if your Nami - itch hasn't been satisfied yet...

>>2057
>It could have to do with size/shape also. I've noticed sagging breasts really want to "weld back" into place, for example.

Yeah that makes sense. Her breasts are imo especially saggy albeit full in most gens from that Lora I've used for Nami. I'm also having similar problems with the Lora I'm currently testing for Jessica from Final Fantasy. Only here the problems stems more from the hardcoded high rise decollete instead of sag.

>Have you tried other samplers like Karras over Euler A?

Not sure if that question was for me or for the guy(s) having trouble to get the Lora working at all, but yes,, I'm using Karras. I've been using nothing but DPM++ 2M SDE Karras for over a year by now. Haven't really kept up to date in that department. Only tried DPM++ 3M SDE but somehow got worse results. Allthough I've not tried other samplers with pony yet. any recommendations?


<=== speaking of well over a year ago; Like my first post in this thread, I've made this Carl redo/remaster/tribute/whatever-you-want-to-call-it last year when dolcettish was still around...

And now I've done a remake of the remake with pony. ...Just amazing how far AI has come in a year!

I know this could have been even further improved, but combined with the time spent when I initially made this I already spent more time on this than on any other single image. ...and enough is enough!

And the last one for now.

This char didn't even need a Lora and just prompting "goblin" yielded a ton of good multi subject pics with tons of relatively consistent swords. I've seen her before but had no idea who she was exactly. I just blatantly copy-pasted the prompt from somewhere on civit and added the debreast parts to the prompt. But judging by what little I read about her since then I guess this kind of general scene (minus the gore) must have been in the show (?) and therefor also in the training data for pony (maybe?)

Definitely worth checking out. Especially since that new anytest thing for controlnet makes it so easy to alter the pics while keeping the overall composition. It also helps tremendously with detail inpainting. I doubt I could have made the stumps etc. look as good as they do here without it!

More to come soon

>>2124
>>2125
These are awesome

>>2126
>Not sure if that question was for me or for the guy(s) having trouble to get the Lora working at all, but yes,, I'm using Karras. I've been using nothing but DPM++ 2M SDE Karras for over a year by now. Haven't really kept up to date in that department. Only tried DPM++ 3M SDE but somehow got worse results. Allthough I've not tried other samplers with pony yet. any recommendations?

Samplers are a bit complicated to discuss because they largely don't affect gen quality, it's just the denoising algorithm.

I don't wanna send you down a rabbit hole, but there are specific situations where switching the sampler helps or hurts.
I don't know all samplers and their behavior, but I know these 3 are good to cycle around:
Euler
Euler A
Karras

Euler A is considered pony's "default", as recommended by its creator. It's good all around, the algorithm "zones into" the target gen but doesn't converge, this means, to name one thing, increasing steps over 30 is worthless, since it will keep slightly changing things back and forth, oscilating over the target, so to speak).

Euler is like Euler A but supposedly better (although theory isn't practice). Unlike Euler it converges. This means your gen will get more "precise" the more steps it has.

Both Euler and Euler A calculate composition during the first steps and polish details on the latter ones. This mean if it decides there's a ball shaped object, it won't change shape after the first steps, then it likely won't change what it thinks the object is, and during the last steps it will barely change detail at all.

So for example, using keyword blend as: [watermelon:basketball:0.8] won't do much at all on Euler, and you'll end up with a watermelon, maybe with some vague hint of basketball lines.


Karras, in contrast, has the unique feature of feeding noise into the gen inbetween steps. This can be a good thing or a bad thing. Since it means it can correct stuff at random by "renoising" random areas, or screw everything up by destroying instead of fixing.
If your prompt is full of contradictory tags, or is just really long, chances are Karras will just fuck things up more often than not. It can for example add random limbs or body parts.

But the reason Karras can be better with this lora is because this renoising helps to breakthe "welding" process where for example, if a severed head is drawn a foot over the neck, Euler would slooowly try to reconnect it, either by bringing the torso up, head down, or elongating the neck.

I thoroughly recommend that you lower preview quality and make it refresh every 2 steps or so, so you can visualize how denoising works. It's a much better explanation than I could ever give.

>>2127
What the fuck
U made this with pony ai?

>>2134

That is a really nice comprehensive explanation. It really explaines a lot of the AI's behavior I've seen. Thank you :)

>Euler is like Euler A but supposedly better (although theory isn't practice). Unlike Euler it converges.

Sry, do you mean Euler A is supposed to be better or Euler?

>I thoroughly recommend that you lower preview quality and make it refresh every 2 steps or so, so you can visualize how denoising works.

Great Idea, I'll give that a try! :)

In general my problem with Euler and Euler A back when I first compared them a little, was that they both seemed to make most gens oversaturated and much less diversified (for lack of a better word), meaning they all looked too similar to each other for my taste. But now I wonder if using Euler/Euler A for HighResFix, Adetailer and some parts of the inpainting process can be beneficial. I'll have to experiment with that. :)

>2151

Uh, well the last iteration was made with pony, yes. ^^ But up to >>2126 it was done with SD1.5 (mostly with wondermix by zovya if I remember correctly.)

<=== here you can kind of see the evolution of that pic. on the left is the original drawing by Carl, the second image is the crudely photoshopped and colorized version that I started the inpainting process on. The third pic then was done after around 4 or 5 evenings using inpaint. Piece by piece without controlnet (which either wasn't a thing yet back then or I just wasn't aware of it, I don't remember. The last pic was then made last week with Pony (icatcher semirealistic), the debreasting Loras and controlnet (Softedge and/or canny) Again some details needed tweaking with inpaint but I didn't spent nearly as long on it than I spent on the third version in this image.

It's a shame really, to see so many real artists hate on AI instead of embracing it (Although there are at least a few). I mean, if a hack fraud like me can turn a sketch like that into something like >2127, then imagine what the guys with the real talents could do! :)

>>2156
"Euler A" stands for Euler Ancestral, Euler is supposed to be better because it converges better.

>was that they both seemed to make most gens oversaturated
That shouldn't have to do with the sampler, but if you're getting that, lower CFG
>much less diversified
This is basically russian roulette depending on your setup, loras and prompt.
For example, if I'm using a strong lora that dominates composition (for example a pose lora with small dataset), then Euler will stick to that, but because Karras feeds noise, it may translate as more varied, because it doesn't "plan ahead" as euler does.

Personally I stick with Euler A. If I get composition locked I either switch the size or rotate the sampler.

>>2128

don't worry, at least one user (me) here knows who she is and appreciates the image.

>>2157

>That shouldn't have to do with the sampler, but if you're getting that, lower CFG

Ah, yes that makes sense. I did those tests back when I still noobishly used the default 7.5.

I've compared the samplers a little again now and my first impression is that the gens with euler are a bit clearer even at lower steps. Dumb question most likely: Are the samplers roughly converging at around the the same amount of steps with pony as they do with SD1.5?

>>2165

Thanks, much appreciated! :)

...and for those who don't like my attempt at humor ^^

>>2205
i like the little stories added to the pics

>>2205
I, for one, applaud the stories AND the pics!

>>2025

No, I'm not the LoRa creator, but I was the OP of the thread.

I just suck at this. Even with the LoRa running, I can't seem to get the AI to actually do any sort of breast separation. Perhaps my prompts are bad.

As for the UI, I'm using Easy Diffusion. I know, some say it's low powered, but I don't see how, it has all the features I should be using (controlnet, etc). I guess I just need to practice.

>>2247

Hmm, at first sight easy diffusion seems to handle emphasis very differently. The readme on the git page isn't very clear about this (to me at least), but if I understand it correctly you'd supposedly have to prompt either: "debreast_fly+++++" or "(debreast_fly)1.5" to get the same effect (or at least weight in theory) as (debreast_fly:1.5) in a1111 and co. Do you use either of these two syntaxes? Does the Lora do anything at all? At least some mutilation and blood?

Also, does the Lora weight handler in easy diffusion even allow for Lora weights above 1.0?

Can you post your prompt or better yet maybe upload an example pic as .png (with prompt in meta)?

>>2213, >>2233

Thanks! In that case there's more to come soon. :)

<=== This next one here though is just the pic; she is most likely just a dumb onlyfans whore anyways... One that probably shouldn't have scammed her followers with old reposts of softcore advertised as hardcore... or something like that...


>>2249

I've read up a tiny bit more about the current version of easy diffusion. I'm not a hundred percent sure, so correct me if I'm wrong, but it looks to me as if easy diffusion does not support/have adetailer. I know how much it sucks to have to learn a new program when you're used to the one you've already got. But the lack of adetailer alone is something that makes me want to highly recommend you consider switching. Let alone the fact it won't let you just use the Lora properly. (I know, there is face restore, but that is absolutely inferior quality wise compared to adetailer imho. And adetailer isn't just for faces...)

I've started with easy diffusion, too, but switched to a1111 more than a year ago, because at the time it only supported Lora but not Lycoris. I dreaded having to seemingly relearn everything but the jump in complexity wasn't nearly as bad as anticipated.

Just a few days ago anon from >>2087 convinced me to switch once more, this time to Forge. Again, at first I didn't want to deal with the hassle of installing and setting it all up, but since there's a 1-click-installer it turned out to be as easy as can be. The ui is almost 100% identical to a1111, so many (if not most) infos concerning the use of a1111 you'll find online apply to Forge, too. If you have enough diskspace (around 20 gb) you could also just install it parallel to your easy diffusion. (you can easily set Forge up to use custom model-/Lora- folders, so you can keep all those files right where you've got them now)

In case you want to give it a go: https://github.com/lllyasviel/stable-diffusion-webui-forge/tree/dev2

Ignore the threatening looking github -list on top of that page and just scroll down until you get to the section "Installating Forge" and click on ">>> Click Here to Download One-Click Package<<<". Aside from that you only need to read and follow the two simple instruction sentences following after the link. Install took only a couple of minutes and didn't need any interventions.

So yeah, to quote that anon who convinced me: >Give it a go, nothing to lose

The Lord works in mysterious ways...

...he also comes to those who can't (or don'want to) read.

>>2294
I freaking LOVE these.

>>2341

Thanks! :)

Allthough I have to admit that the lack of more responses makes me question if these really are ok. Not sure if I should keep posting tbh. Don't want to flood the thread with stuff allmost nobody likes in the end. >.<

That said, I can't be (almost) the only guy posting stuff in this thread. Doesn't anybody else have some nice severed titties to show off? Please? :o

>>2375
They're really good, I can appreciate a well detailed gen with a rich background

I gotta ask though, your upscaling method seems to be adding too much detail on the details if that makes sense, but the overall shape remains a bit melty when zoomed in.
I think it's better to have sub 4k resolution with flawless detail, if you agree maybe I can help, What's your current upscale method?
And please don't take it the wrong way, your gens are already among the best here.

>>2377

Thank you! Sure, if there's a way to improve the upscaling I'd love to know! :D

My current method has two steps if you will (and for cempleteness' sake) First when I have found the right initial gen I do a highres + adetailer pass. High res with ScuNet_PSNR at 0.24 denoise, 16steps, "upscale by" set to the lowest possible size that results in an even numberes resolution (so in most cases upscale by 1.25times). Then after all manips and inpaintings are done I send the image to the extra tab and scale up x4 with "4x-Ultrasharp" (custom upscale model not included in a1111/forge by default) and GFPGAN at 0.300 visibility.

>>2379
I'm the anon from the other thread.
Try multidiffusers, it's what I use and it may be a game changer for you, since you're also having issues with blurry stuff (from what I understood)

Copypasted from some online guide:
----------------------------
>1.Gen your image in txt2img

>2.Send image to Extra tab and upscale it by 2x using your choice of upscaler (I use 4x_NMKD-Siax_200k)

>3. Go back to the txt2img tab while its upscaling and send the original image to img2img (we want the prompt and settings info)

>4.Go back to Extra tab and send the upscaled image to img2img now, replacing the old image

>5.Enable MultiDiffusion, Set it to Mixture of Diffusers

>6.Set the Tile Width/Height to the original image dimensions (ex. if you gen at 1024x1024, that is what you put here, do not use the upscaled dimension from Extra tab)

>7.Scroll to Denoising Strength and set it to 0.33 (anywhere between .25 to .45 is good, higher numbers results in more drastic changes)

>8. Click generate until you get an improved image you like

>9. If you like a gen but only want to slightly change the results then save the seed and alter the Denosing Strength
----------------------------

Mind you, you can still get odd mis-generations here like what you get with inpaint if the denoise is too high.
To get max detail on complicated unruly gens I XYZ plot for denoise at 0.2,0.3,0.4,0.5 .
Then composite the best parts from highest to lowest. This minmaxes clean detail.

But for now stick to the guide, just know
higher denoise = better detail but riskier
Lower denoise = safer but less detailed.

Do go ahead and give it a go, nothing to lose ;)

>>2381

Ok, I'll give this a shot with the next projects. :) But just so I understand this correctly; I should use this method instead of HighRes and then do all the manips/inpaint with NO further upscaling?

>(I use 4x_NMKD-Siax_200k)

was that still a quote from the guide or is that in fact the upscaler you're using?

>>2401
>But just so I understand this correctly; I should use this method instead of HighRes and then do all the manips/inpaint with NO further upscaling?
Yup, you gen without HighRes and only do the guide for a specific gen you like. In my opinion this is faster anyway since you don't wanna highres process trash gens anyway.

I do the inpainting over the upscaled-multidiffused version afterwards.
You can also upscale > correct small things on photoshop and then process on multidiffusion, because multidiffusion works sort of like a full image inpaint, but don't worry about workflow tricks for now, just test it to check the quality gap.

>was that still a quote from the guide or is that in fact the upscaler you're using?
That's from the guide, x_NMKD-Siax_200k and 4x_foolhardy_Remacri are good.
Siax slightly washes out colors so I use Foolhardy. Either works but it doesn't matter that much since multidiffusers will reconstruct detail anyway. Those have the least quality loss from what I've tested

>>2406

>In my opinion this is faster anyway since you don't wanna highres process trash gens anyway

Oh, god heavens, of course not. I only use(d) Highres & Adetailer once I had one specific seed and prompt that I wanted to refine, not for every gen.

>You can also upscale > correct small things on photoshop and then process on multidiffusion, because multidiffusion works sort of like a full image inpaint

that's a neat trick that'll come in handy I believe! :D

Damn, Foolhardy is really good! Siax already was a huge quality jump but this is even better. And not just via multifdiffusion, but as well when used for the regular old way in the extra-tab. Guess I'll also have to redo the upscaling for my Lora trainingdata. xD

>>1865
Hi Eel,can you do a new thread for ass cheeks getting cut off with anything you can come up with? There's not a lot of images and videos of it.

>>2417

Hmmm, I may incorporate something like that in some upcoming works either here in conjunction with debreasting or in the pussystab thread, not sure yet. To be honest, ass cheek ampution isn't something I'm really into that much so I don't think the one or two images I might come up with would warrant a whole new thread. Come to think of it, I can't recall ever seeing a pic of that nature other than crude b/w sketches perhaps. You got any good (preferably colorized) reference pics on pixiv or so that you could link, by chance? I'm struggling to imagine how a somewhat decent and vaguely anatomically correct 'ass-cheekstump'(?) would look like.

>>2494
Might have to back up one page to see the image. 382/669

>>2496

Yeah as soon as I saw the first pic I remembered, I've seen this waaaay back when the guy first uploaded these to pixiv. Hmmm, just bloody muscle tissue then? I thought there'd have to be more stuff, like layers of ligaments, fatty tissue and perhaps even pelvic bones involved. But maybe I was just overthinking it. Allright, I think this should be doable. But please be patient. Like I said in the other thread, I've got quite a backlog right now (like for instance I still got to make the promissed "happy girl cutting herself from pussy to sternum" -pic in the pussy stabbing thread and the 3rd Nami pic for this thread)

Also learning to handle/experimenting with all the new workflow stuff I've just learned here has slowed me down quite a bit.

Any character- or setting- preferences for when I finally get around to your request? ^^

>>2112 hey, btw. I've replied to your message on pixiv, in case you missed it. :)

>>2375
>That said, I can't be (almost) the only guy posting stuff in this thread. Doesn't anybody else have some nice severed titties to show off? Please? :o


Honestly it might be the combination of most threads on //ai/ belonging to a single artist and other not wanting to intefere, and (in my case) not beng a fan of loli.


If anyone have a a good debreasting image i can use a tempalte for Anytest?

My current experiments (picrelated) are not very good

>>2536

>Honestly it might be the combination of most threads on //ai/ belonging to a single artist and other not wanting to intefere

Could be, even though just one look at the op's 1st post should clarify that.

>If anyone have a a good debreasting image i can use a tempalte for Anytest?

Isn't there already something here you could use? Or what exactly would you need?

Multi diffusion is amazing!

...they should have read the fine print in their contracts!...

>>2696
I love these so much! MORE, PLEASE!

Somebody just dropped this: https://civitai.com/models/591887?modelVersionId=660968

was working for a long time on a lora for debreasting. As I am not a huge fan of anime style and original images were anime style had to train multiple loras recursively on the gens of previous ones. Here are some examples of the last iteration

have paired one of the iterations with nipple penetration lora and now I have a new fetish

>>2875
These are delightful! Very well done, and I look forward to seeing more!

I love the ones where her fucks the tits wound,

can you do Hinata?

>>2880
couldn't get hinata lora to work in previous checkpoint, so here is in autism

>>2880
here you go

love it. Thanks

>>2862
very close but never exactly what i want.


thanks for sharing!

>>2928

Works fine on my end. I'd recommend using something like DPM++ 2M SDE Karras instead of EulerA though. I also have had less success with EulerA. Also just as with the debreasting Loras I had to increase the weights for the Lora itself and the trigger words to around :1.5. Still wondering about weights in comfy as there are no apparent values in your images meta.

>>2873
Can someone please do more debreasting pics of Elsa?
Would also like to see her slaughtered like an animal <3

>>2872
>>2873
>>2874
>>2875
>>2876
>>2877
>>2881
>>2883

Pleased beyond measure that someone finally figured out how to do this. I'll get there someday.

Hello, I love your debreasting AI artwork very much, it's so hard to find anatomically accurate debreasting AI pictures. Could you please do a request of Faye Valentine got her breast being cut off, and then fuck the breast wound? Nipple cutting would also be nice. That would be so hot!

bump

Hi any time line on some butt cheeks getting sliced off

>>2944
>>3290

I'm working on a short multipanel thing with Anna and Elsa for now. Will contain (but not limited to) quad amputation, cannibalism, pussy stabbing, debreasting, de-ass-cheeking(?). But may take a few weeks. Not very motivated atm after my last project flopped tbh. When it's done it'll be over in the pussystabbing thread.

>>2973

Just to clarify, this thread isn't mine (or any other contributors exclusive thread), not all pics here are mine, so I'm not sure if your request was directed at me or not(?).

>>2872

Did you ever finish that Lora? And if so where can we find it?

>>3290

Seconded. That would be something to see. Given that there's very little material out there to train a LoRa with.

>>3290
>>3426

Allright, that Anne&Elsa thing is taking too long. 2 female chars + males is hard. So in the meantime I've slapped together this here for you...

...off with her goodies!

...how's that?

But wait, there's more: >>3452

>>3466
I'm hungry for a good a breast burger.

>>3468
Thanks for doing the ass slicing. Noticing that half of her ass cheeks were sliced off, when you do another one take the rest of the cheeks off in one slice. Looking at the after cut Noticing the spot were her ass cheeks are at the back there's a v at the top of her ass, that should be gone and her anus will be visible.

>>3477

Like this?

Deux petits Filets Derrière?

Sry if you had hoped for a larger anus, but I really don't like fat prolapse-y assholes.

>>3511
Yes, right on. The anus is a perfect size for her ass. Thank you 😊

Absolutely beautiful!

>>3470


this kind of Burger?

>>3470

...or this kind?

(you can find more of the first kind on civit on the Lora's page. Just search for "breast food")

>>3515

Glad you like it. :)

>>3517

thx! :)

I had wanted to show this classic Carl pic some love for quite some time. (original for comparison: https://e-hentai.org/s/447a39826a/72277-6) ...Never got around to it until now.

Bit sad that dolcettish is down. There was a thread where Carl himself posted, not only his original works, but also some AI-reworks someone made on pixiv. Would've loved to know his opinion on this humble little attempt here, but I doubt he frequents this little corner of gurochan these days. In case I'm wrong with that assumption: Hi Carl, we miss you! Do you still post somewhere? :)

P.S. I know this isn't strictly a DE-breasting, but I think it still best fits in this thread here.

Bump.