>>1178
>the funny part is that I could easily see the dump thing happening in PRAD.
I can't, what episode plot would involve the twins hanging out in garbage?

>>1179
Not an episode specifically, I mean the girls have wacky hijinks at the dump and the twins show up

>>1180
rhythm eats a weird piece of meat at the dump

That sounds like a pripara bit, but it works for PR

I am looking forward to pripara and kind of annoyed how the experience of watching dear my future and rainbow live is getting into a new group of girls for 50 episodes then they get dropped

>>1159
>>1160
>>1161
>>1162
This company is more powerful than most governments, by the way. What a world we live in

>>1184
Even though they get regulated regularly and are consistently seen as incompetent on media...

They're not even like Samsung who owns half of South Korea and all the government

>give anons the power to make anything with AI
>they make chubby girls and futa
grim

>>1187
>grim
green

>>1105
I decided to give Pony another try, or more specifically I checked out a finetune of it called AutismMix and it seems quite impressive. It can even do sex positions! There's still errors that pop up, but like AI in general the reason it works is because your brain is turned off when fapping. The Japanese character recognition is mediocre (Reimu works but Sanae doesn't??) but obviously still far better than my own merges that are like 80% furry just so genitals can be generated. I still find it funny that I knew the furry connection within a few weeks and it took other people over a year to notice it. Furries are powerful.
I really don't know how to prompt for it, but I guess I'll learn eventually. Pic related is what it looks like when I try to prompt Kuon (with other tag assistance) and it completely lacks her beauty and charm of course. Unlike what I previously thought, you can train LORAs with even 8gb of VRAM, so my 12 will allow me to make my Kuon and Aquaplus LORAs again, but I have to do the image cropping all over again because it's 1024x1024 instead of 512x512. Soon...

I'm still going to keep my old models around, not just because of the hundreds of LORAs I have that are not compatible with SDXL, but because I like the general style that my merges have. I may try making SDXL/Pony merges, but I'll see how things go first. It seems to have less options when doing the Supermerge process so that may make it easier.
In other news Stable Diffusion came out today (or will very soon) but like all the other AI stuff I don't have any interest until someone makes a 2D model of it.

>>1189
>In other news Stable Diffusion came out today
Err Stable Diffusion 3 that is

I was looking around 4chan today and happened to stumble upon this thread, https://boards.4chan.org/a/thread/268171362

It got me kind of curious because from what I know most models when you try to edit an image directly tend to alter the base image a little bit into something else instead of perfectly editing the image with a specific modification. From what the OP said in a recent post he's using some sort of subscription, but I know that DallE and Gemini don't work like this so it has to be someone's paid SD offshoot that they've tweaked to work in this way. My question is how would you approach doing this in your own SD model? Via controlnet or something? It seems so odd... Of course there's still plenty of areas where it's making unwanted changes like with changing the style of the bras and aspect ratio or whatnot, but for an advancement in AI modifying only specific details of an image it looks like it's doing pretty good.

>>1191
Looks like some skilful usage of inpainting.

>>1191
He's just using NAI probably, although I didn't know they offered such a thing.
It's inpainting, yeah. I don't really do it much, but these days you can use controlnet which would probably be superior to NAI's, although they might have their own model for it or something.
You go to the Img2Img tab and then the Inpaint tab, provide the prompt, and ideally set up controlnet. You use 'inpaint_global_harmonious" and set denoise to 1 instead of the usual 0.3 or whatever.

remember how image generating AIs actually were around for years and years before stable diffusion took off and how janky they were

>>1761
Yeah, it's funny how long AI stayed in that relatively Coming Soon™ state for over a decade until one day it all just blew up and insane progress occurred nearly overnight.

There's a Chinese local video model that came out recently, but right now it's limited to text-to-video. They've said there's going to be an image-to-video one released in January, which would be the one that's far more fun since a generic video model isn't going to know 2D stuff... I think?
This is a situation where 24GB of VRAM like on a 3090 or 4080+ would turn video creation from 20 minutes into 20 seconds, so it sucks to not have one of those. Currently it seems like you need the "ComfyUI" setup instead of the usual automatic1111 SD UI, which is also unfortunate. It must be a joke name because nothing about its UI is comfortable.

https://x.com/STOP_THINKINGAA
Jesus Christ...

At some point you have to wonder how much morals weigh out against just being a scumbag and taking full advantage of ai, this guy's making 2.6k a month doing this. And looking at his 18+ version of the account, https://x.com/HachimidesR18, it's easily visible how it's completely automated. No thought or care put into anything just the same exact poses with a pallet swap. I don't understand why anyone pays for this when they could easily just start up SD on their own and do the exact same thing.

>>2213
I don't know how people could possibly pay for AI stuff. I will say at least that he must spend a lot of time cleaning up the images. It lacks creativity or personal input so it's the most inane stuff you could imagine and it's why I never do it. It's not quite manual labor, but it's very dull and repetitive.
The part that bothers me is that an AI generator person's "style" is just a combination of real artist tags, which adds insult to injury when people expect (and do) get paid for it. My opinion is that AI really shouldn't be used for anything other than a personal masturbation aid, which it is very VERY good at. Well, and making throwaway images that people look at for a few seconds to laugh.
The only paid AI stuff that could possibly make sense to me is the extremely niche stuff that has little no real art.

>>2213
It's good because it's novel

>>2213
People are dumb and easy to take advantage of. I'd be a multi-millionaire if I didn't have good morals. The internet is designed now so that these people rise to the top. It's why youtube is filled with dumb people making thousands a week stealing content from people that actually put in effort and know what they're doing.

In the last year alone I've have multiple aspiring vtubers take content I released for free and shit all over it. All because they're trying to get sponsored by Funimation/CR. So they take something we worked hard on and lie about it. Last time they took something with multiple TL notes and subtitles at the top of the screen and hid them. Then presented the subtitles at the bottom of the screen in such a way that they somehow made the offical (wrong) translation look better. In that particular scene we provided two translations. One that attempted to maintain the joke being made and another literal translation. Maybe we could have handled it better. But the one that retained the joke was kind of iffy and it was really hard to do that joke in English. So we put it at the top of the screen and a literal translation at the bottom.

In another scene they either lied or made a common mistake. Obviously they don't know Japanese. But they used it as a way to claim the CR TL was better. Since whoever did it didn't understand Japanese either.

This kind of thing isn't just limited to people trying to make youtube money either. We've had multiple people try to shut us down on places like nyaa. If you dig into who's running the place now you discover the same thing: It's all CR or former CR employees. That's why they push official translations and rips so hard these days. They're the same ones that ddos'd all the anon nyaa clones as well (after they stole the code anon provided for free to run their own).

It's a really horrible situation now on the internet. People that actually do good work are shut out of everywhere with gatekeeping. Any attempt to start up a place outside of these people's control get ddos'd into oblivion and if that fails they email your hosting provider or domain provider and shut you down that way. They steal your work, claim it as their own and then get donations from clueless people that praise them for their "hard work". These people using AI are just the same thing happening with art.

Why do people pay them and support them? No idea. I'm going to guess they're lonely and any interaction they get from a "virtual friend" brightens their day. Things are so bad we've monetized friendship. It's really sad.

>>2219
Thank you for your hard work.

https://huggingface.co/deepseek-ai/Janus-Pro-7B

Anyone tried this yet?

>>2333
These models don't matter to me until someone trains it on booru stuff to make it palatable. There's like 50 different real life models around and I can't care enough to compare them. "Oh, that one does apples on tables better than the other one which is better at mountains."

After installing a new OS I was reminded to disable hardware acceleration in browser because at my 12GB I can't generate on SDXL with browsers eating up VRAM. I couldn't figure out why it kept hanging at the 98% mark, but then I remembered that I am barely squeaking by.
I've moved onto illustrious/noobAI models, somewhat. I made a merge combining 4 different models and I'm somewhat happy with it and I'll throw it on mega if anyone wants to try it. I could throw it on civitai maybe to get some credits if people like it, although I should check permissions to see if that's okay with the model creators.

Illustrious/NoobAI aren't as good as pony in sex stuff, but quite good at "1girling" with characters and artist styles. It can generate a lot of characters and artist styles with a simple prompt, no lora needed (although LORAs are always better, just more work).
Older models would certainly not be able to handle "bird perched on hand" as it's not even a booru tag. Oh, and of course it clearly recognizes Kuon which is amazing in its own right!

>>2762
It's on mega now if anyone wants to mess with it locally instead of using kissu's prompting thing in /chat/ or irc (still trying to iron out the issue with the sampler there):
https://mega.nz/file/abw3jSTA#7fYEMUeF0qR07jj3h9cAgp8YNhKTI6zeDERJSBUNAfM
It's prone to making a splotchy appearance if you don't use the right 'Schedule Type', and it seems like Align Your Steps is the best at it, but I'm not sure since I haven't done a lot of testing yet.
I should do more merging tweaking, but I don't want to spend too much time on AI when I'm supposed to be doing 3D modeling stuff that's actual creative work.

>>1191
Back in my day we had to DRAW boobie edits BY HAND!

>>1191
this just came out today but I haven't tried it, looks like it only works with comfyUI for now because everything else uses white masks instead of black masks apparently but surely someone will fix that soon
https://civitai.com/models/1376234/noobai-inpainting-controlnet

https://www.unite.ai/civitai-tightens-deepfake-rules-under-pressure-from-mastercard-and-visa/
https://civitai.com/articles/13632

Our credit card overlords have decided civit.ai needs to make some changes. Currently it's for real life celebrities so it doesn't affect anything I care about, but once their claws are in a target they're not going to release them. They're going after civit.ai's monetization which is its ability to stay online so it might not be too crazy to start hoarding things in the near future.

Good lord. ComfyUI is... AWFUL. I can appreciate the tongue-in-cheek name, but that's about it. Sadly most stuff released in the past year or so is aimed at this thing because the older UIs stopped getting big updates. If you want to try generating any of the new video stuff, you unfortunately have to do it here.
This is definitely a UI made for programmers because it makes my eyes bleed. Someone looked at texture stuff for 3D models and decided "Hey, usable UIs are too good, let's replicate that for image generation!"

I'm trying out some ComfyUI workflows. I think ComfyUI is the solution to SillyTavern integration if you can get some sort of text interpretation model loaded to take chat text and convert it into booru tags. Or maybe it's something that should be done in ST itself?
Well, whatever. Either way, ComfyUI does indeed generate faster, although I'm not sure why ADetailer is taking so long here.

Oh, and after about a week of looking at giant comparison images I have the Kissumix 2025 FINAL model made, assembled from the layers of like 8 other models. I'll upload it later. Unless something drastic happens it's the last SD model I make in 2025. I'm not sure what else could happen with the tech.

Well, this seems to work at preserving the style. It's a looping video model called JACE and I have no idea what it stands for, if anything. It takes an image and uses it as the starting and ending frame. It's combined with a checkpoint based on "looping wallpapers" ( https://civitai.com/models/1655054 ) and it seems to work well, but I'll have to try it with other models to see if I can make it more animated.
This will work well for SillyTavern portraits, though. Definitely no 'live' generation since it takes a minute to generate each second with somewhat exponential generation time.

hehehehe

>The woman in the first image is eating a popsicle. She stands to the left of the woman in the second image. The background is the same as second image.

Oh man this stuff is good, and all done locally on my computer. The prompting is very annoying, though.

>>4277
I can foresee the power of this technology in the practical application of hentai doujinshi edits.

>>4277
>>4278
This is actually really cool for speeding up edits.

>>4277
>>4278
Sugoooi... More Koupen, please!!

was trying to get the #chatgpt kissu bot to run on my machine

>>2762
nice bird!

>>4315
>>>/ec/ !

Good lord what is with this workflow. It seems complex, but surely it doesn't need to be THIS convoluted.
Well, time to see if I can get this working for WAN 2.2 video.

Wow, Kuon is COOL!
(and I forgot to change the included prompt)

>>4748
and man... all these models are always talking. It seems impossible to stop the mouth movement...

>>4747
>>4748
I see AI interfaces have taken inspiration from modular synthesizers.

Have you seen those videos from the so-called slop thread linked in Happenings? Some looked very impressive with how they displayed a realistic framerate and less of those L2D-ish movements.

>>4750
>Some looked very impressive with how they displayed a realistic framerate and less of those L2D-ish movements.
I'm still in the process of trying to mentally decode this confusing mess of a workflow. I already know how to do better animations, but not in this new workflow thing with WAN 2.2.

>>4750
My guess is like hand drawn animation we're going to see digital drawn stuff go away in the near future. There is no reason to employ so many animators at slave wages and deal with out sources (and the bad PR that comes with it) when you can produce an entire show using nothing but prompts and some basic NLE.

Embed is already 2 years old and look at how positive the reception to it is. Now imagine doing the same thing with today's LLMs trained on more data and more effort put into the story.

The only saving grace is maybe this will allow someone that couldn't produce something otherwise to give us an amazingly epic show. But for every one of those there will be over 9,000 horrible productions made for no reason other than ad money. All of them will probably drown out anything good because the ad revenue systems favor clicks and quick fire garbage rage bait.

People will watch it. I've been around long enough to see how most people will accept the decline of effort in entertainment. I was around to see television switch over from telenovels, sitcoms and series over to reality TV. I saw cartoons and anime go from high effort hand drawn stories and animation to low budget CGI and lazy story telling. The decline is pretty obvious. Plus now there is less chance for something interesting to be aired on some forgotten channel. Even with the internet we have less and less outlets for people's work to be seen with each passing year. 1980s cable television had more unique and hidden stuff than the modern internet. As unbelievable as that might sound. Things don't get shared around the internet now like they used to either. Everyone is following the same handful of trending lists and not sharing or seeing much outside of them.

I'm all for live and let live. But as far as art is concerned things have been on the decline for a long time. This holds true for stuff like oil paintings as well. I'm sure someone somewhere is still making masterpieces but good luck seeing them shown anywhere.

>>4751
Seeing this brings back memories of how I felt when I first learned Adobe workflow, later on I was glad for the hundreds of options and combinations you can achieve. You have my blessings that it will be the same there.

>>4752
I see it the same exact way for the things I was able to see change over time. The style of the video you linked brought an example to my mind: Have you ever seen action scenes from the garbage that is Netflixvania? The framerate dips down to below 10 or so and any will to cover up the obvious rotoscoping goes completely out the window, yet a worryingly large number of people for some reason mistakenly believe there was an extraordinary amount of effort put into the segment and that - in their words - "they put half the budget into that scene", as if they had just seen a masterclass in animation. It's absolutely insane. I'm not denying that traditionally animated shows from the 90s used rotoscoping as well, but at the very least they knew how to make it blend in or used it more as a reference to keep the visual flow. Even the heavy reuse of stills with moving mouths was incorporated in a way where it didn't feel odd, certainly not on the first watch at least, unless you're going back to older productions like Filmation's He-Man and She-Ra. One of my biggest disappointments is with Big O and how absolutely atrocious its second season looks where they switched to digital animation, though even that still looks better than a lot of things over 20 years later.

Either way, to get back on topic: I personally don't really mind (anymore), I rarely watch modern things even if they're of interest to me, simply because I don't like digital animation, including good examples. I still have my favorites I can always fall back to, also in regards to internet platforms, so "AI slop" can't take anything from me if new things I used to be excited for already got replaced in the 2000s. It is due to that that I actually think AI is the most exciting trend since then, a great help in editing that is easy to hide with prior knowledge and, as you said, a tool for otherwise "untalented" people to get their ideas out there for cheap. Autistically obsessed with a certain game genre you have really good ideas for? Use it in tandem with light studying to get something up and running quickly, all assets included, then commission clean-up in all areas if you think you got a great base.

>>4751
Success! Not exactly what I wanted, but this is 4 sets of prompts. If the tendency for it to talk wasn't there, it would look a lot better. I prompted it so that her hands moved to her chest like she was going to reveal her breasts, but then she sticks her tongue out.
10 minutes to generate, but this workflow is made for weaker cards with less VRAM so I should be able to make it faster.
I need to figure out why the animation looks so stiff, like a gacha live2D thing, though. Could be something some LORAs would solve, I guess.