[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [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
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

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]


Janitor acceptance emails will be sent out over the coming weeks. Make sure to check your spam folder!


[Advertise on 4chan]


File: ADT.jpg (3.84 MB, 4569x2025)
3.84 MB JPG
Previous: >>109471633
NEW: https://huggingface.co/circlestone-labs/Anima
Anima1.0 is out

>UIs
ComfyUI:https://github.com/comfyanonymous/ComfyUI
SwarmUI:https://github.com/mcmonkeyprojects/SwarmUI
KritaAI: https://github.com/Acly/krita-ai-diffusion
Forge Classic/Neo: https://github.com/Haoming02/sd-webui-forge-classic

>How to generate anime:
image : https://rentry.org/comfyui_guide_1girl
video: https://rentry.org/wan22ldgguide

>Style Explorers
Anima: https://anima.mooshieblob.com/ | https://animadex.net/
Illustrious: https://tagexplorer.github.io

>Checkpoints, LoRAs, Upscalers, & Workflows
https://civitai.com
https://civitaiarchive.com
https://tensor.art
https://openmodeldb.info
https://www.seaart.ai
https://www.liblib.art/

>Resources:
Share Metadata: https://litterbox.catbox.moe
Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-beta-one
Online metadata viewer SD/NovelAI: https://spell.novelai.dev
Catbox/Metadata Userscript: https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fb
Output cleanup: https://rentry.org/RemovingDiffusionGunk , https://www.mediafire.com/file/vipr23exc5htmnt (batch processing python script)
Inpainting Guide from an Anon: https://files.catbox.moe/fbzsxb.jpg >>106520607

>/adt/ logo:
https://files.catbox.moe/pakfgt.png

>Beware inorganic activity
https://rentry.org/animanon
>>
Thanks for bakering
>>
File: 00001-2660444572.jpg (1.08 MB, 2048x2048)
1.08 MB JPG
>>
do you do these collages manually or what
>>
>putting every single image in the collage
utter woke nonsense
>>
File: too flat?.jpg (116 KB, 704x1024)
116 KB JPG
>>109497162
participation trophy ass collage
>>
Anima inpainting

Question, does anyone else have trouble with Anima inpainting in ComfyUI? Do you guys use the normal KSampler to do it or route it through another specialized node like LanPaint KSampler?
>>
>>109494151
It's controlnet.
>>
>>109497050
NTA but you can do what he did in like 30 seconds with the right program. The question shouldn't even be asked.
>>
>>109497380
Why not, it's not like everyone knows it.
What's the program, tell me
>>
>>109497390
nta but you can done it quick with ffmpeg & bash script on folder that has the images.
claude/codex can 1-shot it
>>
File: Anima_00809_.png (1.29 MB, 1232x1800)
1.29 MB PNG
Why does Photoshop always feel janky af for text editing. It feels so bad. I edited this with CSP after being frustrated and it was a blessing of UX/UI. Love CSP
>>
>>109497317
Why would you controlnet something that simple?
>>
>>109497419
nta but I still haven't managed to get that exact same pose though
>>
>>109497418
cute ; the vest is well done

>working on dynamic poses and action poses myself
>>
File: 140394037_p3.png (1.17 MB, 1152x2016)
1.17 MB PNG
>>109497419
I controlnet everything to get unique non-ai composition and detail. And it's fun.
>>
>>109497515
nta but thats great the more power you have the better, thats what comfyui was meant after all
>>
File: 144509938_p0.png (2.49 MB, 1608x2072)
2.49 MB PNG
>>109497534
I don't use Comfy. I hate Comfy. Forge Neo is all you need.
>>
>>109497547
>moar feet
>>
>>
>>109497515
And looks nice too
>>
>>109497515
The problem with using controlnet is that whenever you look at the images you generated they’ll remind you of the 3dpd you used as ref
>>
thanks for bakering
>>
>>109498291
>fwesh bakees just for you anon
>>
I don't like the schizo is baking now so no thanks you from me >:( bring rakun back!!!
>>
>>109498475
This. Mentally ill drama freaks should go to /ldg/
>>
>>
We should put anistudio in the OP next time fellow anime gen enjoyers. It's really good and cool and totally well designed and not bad at all!
>>
>>109498501
>doesnt sound suspicious at all
>>
Anyone else using Claude code or other coding agents to help with ComfyUI? I've had decent success with:

>run a docker container with gpu access (to sandbox the coding agent) and tell it to install comfyui and get it running on my gpu. Tell it to only install and manage all dependencies with pixi so everything is reproducible, run everything with "pixi run"
>walk away, it installs some sort of headless browser and tests everything works
>download some workflow from civitai and tell it "get this shit working" and test that its working
>walk away for an hour
>get back and its installed all custom nodes and dependencies and downloaded all the models the workflow referenced
>the workflow doesn't do something that I like
>tell it to fix it

I'm wondering if I even need the premade workflows as a starting point or if just telling it what I want done would be enough.

And yes I realize that having an ai agent install random nodes from the internet isn't exactly safe, but that's pretty much what I was doing on my own anyways so I figure its a wash. I suppose if I wanted I could ask it to review the code for any custom nodes it installs to make sure its not trying to steal anything, but I figure as long as i'm running it in docker I should be good.
>>
File: 156772342.png (675 KB, 832x1216)
675 KB PNG
>>109498630
I haven't used it with a coding agent, only Claude chat

I have sent Claude the JSON from an Anima workflow that I downloaded from Civitai. Turns out it was badly wired and poorly optimized, but I used it as a base to improve it, and then added inpainting nodes as well. I've been super happy with the results so far.

From my experience vibeslopping a 40k+ line project, AI works the best when you give it detailed instructions on what you want and iterate on that result. Avoid abstract and too general orders, unless you wanna get shit as a result/burn tokens like crazy.
>>
File: Anima_00932_.png (1.58 MB, 1232x1800)
1.58 MB PNG
Why is preggo so good bros
>>
File: C057.jpg (212 KB, 768x1152)
212 KB JPG
>>109498784
are you sure you don't have a thing for armpits?
>>
File: 1778133201301445.jpg (430 KB, 2240x1664)
430 KB JPG
>>
>>109498488
We've got NAI V5 testers in the thread!!
>>
>>109498541
that is the schizo pretending to be ani so I hope you know the difference at this point
>>
>>109498501
Isn't AniStudio malware?
>>
>>109498501
>>109499079
You really should kill yourself now, Julien
>>
>>109499198
>>109499216
getting found out then doubling down is always a losing strat
>>
>>109499227
Speaking from experience, eh?
How much did the japan funding run raise, a negative amount?
>>
>>109499198
yes
>>
File: Anima_00946_.png (1.52 MB, 1232x1800)
1.52 MB PNG
>>109498898
lol it's just that pose but theyre good too
>>
>>109499198
It's an .exe file, duh.
Obviously...
>>
File: ComfyUI_00005_.jpg (875 KB, 1433x2150)
875 KB JPG
Anyone know when the checkpoint for krea2 with danbooru tags is coming?

I tried recreating an old gen I did in anima in krea2 and the difference is night and day
>>
This was anima and I had to use regional prompting otherwise characters with similar features were bleeding together
>>
>>109500012
there is a kroma lora that lodestone made with a bunch of danbooru and e261
>>
>>109500132
I tested it, it's shit
>>
>>109500192
ty for testing. other than that there is the Lora to turn it into an edit model which might work for style transfer
>>
>>109500205
I tried that lora as well, you'll get better results just using klein + anime lora
>>
>>109500012
>>109500021
Tbh I prefer the Anima one, the Krea2 looks extremely AI sloppish, like as if their heads were photoshopped there.

I like the concept though, it's fun
>>
>>109500626
Its more because of my prompting issue than krea2 issue.
>>
>>109500784
krea2 seems smarter but I find it weird with how you juggle the attention to get what you want. anima takes too many retries to land something acceptable. this is in reference to complex prompting tho, not the styles
>>
I'm using minimax r3 default comfyui template and it's shit. I have a 5070ti and I'm reading this rentry article in the OP noob guide for wan2.2 and it says 16gb is enough for the model but that clipping will add more and that 16gb won't be enough?
Am I locked out of wan2.2 with my GPU? I read a little further down and it said I could remove some shit but it will look like ass?
I'm a /g/ tourist and a brainlet
>>
Hey, i've updated my artist tag explorer/ranker, it now supports Anima local diffusion as long as you're using ComfyUI or ForgeNeo, try it if you want!, at this point i mainly use it to automatically queue images of prompts i like.
>>
>>109501002
Im an idiot, the link is.
https://github.com/SeaBassBand/NovelAI-Artist-Ranker
>>
File: 00057-2705079690.png (2.45 MB, 1280x1920)
2.45 MB PNG
What is the most active anime diffusion thread on 4chan?
>>
>>109501101
/ldg/ apparently but one person is spamming most of the anime there
>>
File: ComfyUI_temp_jxyno_00197_.png (1.15 MB, 1024x1024)
1.15 MB PNG
>>109501101
/e/ or /h/ I imagine
>>
File: 1767834828183334.png (1.69 MB, 1024x1024)
1.69 MB PNG
>>
File: file.jpg (294 KB, 1031x1531)
294 KB JPG
>>
File: 00136-1937835226.png (2.18 MB, 1280x1920)
2.18 MB PNG
>>
>>109500952
Modern UI all efficiently offload to RAM what doesn't fit in GPU. It's no big deal. Having 64GB RAM is actually more important or you will be swapping with SSD like crazy until it croaks.
>>
>>109502418
>64GB RAM
Woah excuse me Richie Rich
>>
>>
File: 1764806249227824.png (2.42 MB, 1536x1024)
2.42 MB PNG
>>
File: C092.jpg (216 KB, 896x1152)
216 KB JPG
anime girls
>>
>>109502418
Thanks anon. Still trying to figure this shit out and get wan2 working. I have 32gb ddr5 is that bing chilling? Im doing this shit on my os drive right now. I can get a pic adapter for m.2 but shit is expensive right now-would SATA drives work well or are gonna be a bottle neck
As a side I've gotten better results from minimax with better prompting but still comes out cursed occasionally
>>
Can anyone help me?
I'm still learning and following the 1girl guide.
I'm trying to inpaint and use the sketch tools but somehow it didn't work at all. Masking still works though.
>>
File: random vtuber.png (795 KB, 1024x1024)
795 KB PNG
>krea 2 turbo prompt: 1girl, white hair, fox ears, white jacket, black shorts
>shits out fubuki
kek
>>
Anima is really fell off but my favorite character Lora is on Anima only.
How to convert Anima Lora to Illustrious ?
>>
>>109496881
what model could I make coomer art with for my smut stories? I've been using a site that costs an arm and a leg. I don't know much about AI I just know it's hard to run, all I have is a 7900 xtx. Could I make my coom or should I just stick with spending retarded money for my porn addiction?
>>
File: ComfyUI_00049_.png (934 KB, 688x1192)
934 KB PNG
>>109504362
I downloaded comfy and I've been trying to get the hang of it. it's so fun! idk why I didn't do this sooner! my gens are shit obviously but still, fun stuff!
>>
File: 1775384630127848.png (3.04 MB, 1664x2240)
3.04 MB PNG
>>
>>109505234
Cute
>>
>>
>>109505205
Have fun. what are your stories about?
>>
>>109505693
just random scenarious that pops into my head. It's kinda worst than porn since it's like, perfectly catered to my niche interests lol. but it's fun writing stuff down, and now I can have images to go with them which will give me a lot of reread value I think! they're just for personal use too, I don't post them anywhere.
>>
File: Krea2_turbo_00026_.png (1.45 MB, 1024x1024)
1.45 MB PNG
I'm trying some new models after a few month hyatus, This is made using Krea 2. I'm testing the model 'lore' by judging its knowledge of Touhou characters (and it seems it does a decent job if you give a long enough description of the character the longer the prompt the better)

I'll try famous and more obscure characters. What do you think about it?

In my opinion, this model represents the evolution that Flux should have had (more information on the character and art style).
>>
>>109506109
If you want to test it with 2hus, use nuwhos like Ubame, Yuiman, or Hisami for a bit older.

>>109504362
7900 xtx is more than enough, better than my card and I still get under 10 second gen times.
The first install instructions are daunting but you'll get the hang of it
>>
File: Krea2_turbo_00031_.png (1.42 MB, 1024x1024)
1.42 MB PNG
The prompt adherence is spectacular, even if I notice some anatomical errors in some gens,which is a bummer. Maybe due to the fact that I use the turbo version and an fp8?
>>
File: 1768377092282162.jpg (954 KB, 1904x1904)
954 KB JPG
>>109506137
At the moment the only thing it’s good at is handling long schizo prompts to make gacha tier wallpapers. I still mostly use Anima when I want goon pics
>>
File: Krea2_turbo_00033_.png (1.39 MB, 1024x1024)
1.39 MB PNG
Wakasagihime , i'm more than surprised,but i had to give a REALLY detailed description.
>>
>>109505465
Sex pig
>>
>>109506203
It's better for BDSM and certain related fetishes and is excellent at multiple distinct characters doing different things (also, not coincidentally, an amazing plus for BDSM)
>>
Found this neat repo: https://github.com/SuzumiyaAkizuki/DanbooruSearchOnline
Semantic search for Danbooru, for finding Danbooru tags. You can describe what you want instead of knowing the exact tag name and it'll find related tags for you. Seems useful for SDXL Anima NovelAI prompting when you know what you want but have no idea what the Danbooru tag is, also has related tag and artist search.

Nya nya nya Anonnies /ᐠ - ˕ -マ Ⳋ
>>
>>
File: 1786214307572914.jpg (822 KB, 2976x2104)
822 KB JPG
https://huggingface.co/lodestones/Kroma
Kroma 0.2 is out btw. Looks like he fucked up the turbo version it needs turbo lora to produce any output. Pic related right is turbo, left is base
>>
>>109506800
>both have the halo fucked up
>but right is worse about it
>>
File: 00006-3676744446.png (2.82 MB, 1344x1984)
2.82 MB PNG
AnimaPulse is a very good finetune. Not sloppy, not overburned. Does artists styles pretty accurate. Stable and easy to use.
>>
>>109506908
>AnimaPulse is a very good finetune
only slop merges exist for anima. nobody fine tunes it and no 10k image range isn't good enough
>>
>>109506908
Not bad tbqh. I usually use instant loss 2koma tags to generate a lot of sex actions in a single gen and this handles it quite well, a lot better than WAI-Anima
>>
Nvidia drivers still suck shit on Windows & Linux
>>
What, psychologically, drives people to use shitmixes on anima, a model with a perfectly fine base style, no strong aesthetic bias and really good artist replication?
>>
>>109507124
>no strong aesthetic bias
until you use quality and score tags... then it goes to AOM slop
>>
>>109507124
>really good artist replication
are we using the same model?
>>109507109
works for me
>>
>>109507124
>SAAR kindly add masterpiece best quality absurdres for best quality SAAR!
>>
File: Pasted image (291).png (1.58 MB, 1120x1440)
1.58 MB PNG
>>
>>109507506
Catbox?
Also post the behind view
>>
File: ComfyUI_temp_pdosg_00002_.png (2.31 MB, 1024x1536)
2.31 MB PNG
>>109507124
>perfectly fine base style
Nah, it's definitely worse than shitmixes in terms of overall competence like anatomy, poses, facial details, etc.. Don't get me wrong, I still have Base around since it's most adherent to certain styles and doesn't "insist upon itself" as much as mixes but it definitely has its downsides
>>
I've made a few threads on /a/ over the past week with AI OPs (Minimax webms), even talking about AI in the OP, and they never got deleted by a mod. Even with the usual luddites throwing a tantrum, they still didn't get removed.

It looks like the /a/ mods have now issued a clear directive that anime-related AI is permitted on /a/. Some users will still seethe in them, but they don't decide the rules. Just something to keep in mind if you're an anime genner.
>>
>>109508408
they never did get removed for the longest time. the only fag that gets banned is wanschizo
>>
>>109508432
You know that isn't true, ani.
>>
>>109508440
go to bed schizo, I am not ani
>>
>>109508444
You're not very good at being discreet, ani. We've been through this before, your autistic posting style and labeling of other posters sticks out like a sore thumb. I get that you're not smart, but try a bit harder.
>>
>>109508456
Us mentally healthy people don't give a shit about this ani stuff man, just shut the fuck up.
>>
>>109508496
Relax ani, you already gave yourself away. Don't make it worse for yourself champ.
>>
nothing's a better feeling than seeing two people i hate fight... =D
>>
is julien here? i have a question
>>
File: ComfyUI_01192_.png (1.49 MB, 1024x1024)
1.49 MB PNG
Not nearly enough love for megane. Whatever, more for me.
>>
File: hina.png (659 KB, 653x1568)
659 KB PNG
>>109508940
Here's a meganekko.
>>
File: ComfyUI_01196_.png (1.49 MB, 1024x1024)
1.49 MB PNG
>>109508961
SOMEONE gets it
>>
File: shion.png (475 KB, 399x1218)
475 KB PNG
>>
File: 00027-444141168.png (2.89 MB, 1344x1984)
2.89 MB PNG
>>
File: file.png (1.18 MB, 896x1152)
1.18 MB PNG
>>
nai5 is unet based
Thoughts?
>>
>>109509929
>barely able to write
Back to school monkey
>>
NAI is shit, there are my thoughts.
Why would I want SAAS slop that isn't even as good as local
>>
>>109509929
The thing is, ComfyAnonymous (a literal nobody in finetuning) never would've funded Anima if tdrusell had two brain cells to rub together and just stuck with UNET. Blind funding blind, this is how you get here.
>>
muh 12b hyper bloated half baked anime finetune sirs, muh 20gb checkpoint, muh 10gb text encoder, muh 2gb outdated VAE, muh 5gb censored prompt enhancer, muh cope turbo init convrot quant.
>>
It takes a lot of misdirection but turns out you can get something slightly ecchi from gpt. And the image quality itself is quite nice, prompt adherence is unlike anything out there.
>>
>>109510256
chatgpt*
>>
>>109510256
>And the image quality itself is quite nice
Until you look at those freaky hands or the obsession of chatGPT to add a quadrillion hyperfine "details" where they don't belong
>>
>>109510289
Never said it was the best hehe
>>
File: HO_BMJyWgAA8uUo.jpg (502 KB, 2392x1024)
502 KB JPG
>>109510083
>why would i want a model that can do this as a raw gen
why indeed
>>
>>109510360
Catbox? If you will
>>
>>109510379
got it from twatter, sorry
https://x.com/Timmy_Tommy_1/status/2085097748781846942
>>
>>109510432
Pain
>>
>>109510360
do you notice how they're all facing away and most the hands are hidden(and most of the ones that aren't look fucked up)?
cuz I'm noticin'
>>
>>109510637
stop nyoticing and get excited for new product
>>
>>109510637
could your favorite local model do the same thing better?
>>
>>109510673
>bro never heard of controlnet
>>
>>109510673
doing something poorly isn't much better than not being able to do it. Also, better question, why the fuck would I give a shit about doing a dumb lineup shot anyway?
It's certainly a cool tech demo but it's not worth much in practice.
>>
>>109510700
it's much better than not being able to do it because you can just do a few small inpaints to fix up the broken parts if you can get 90% of the way with a raw gen
>>
>>109510749
v5 isn't launching with an inpaint model(or vibe transfer or character reference)
>>
>>109510764
you can surely inpaint with V4.5 or a local model
>>
>>109510845
keep shifting the goal posts while refusing to answer why anyone would actually want this particular feature, especially if everyone has to be be wooden cutouts facing backwards without any form of character interaction.
>>
>>109509929
Based. It's kind of sad they charge that much for a model no bigger than Anima, but that's how it is.
>>
File: file.png (1.84 MB, 896x1152)
1.84 MB PNG
>>
>>109510888
it's clearly to show off their asses. nobody says they have to be posed like that other than you
>>
File: 1769462991944563.png (2.51 MB, 1536x1024)
2.51 MB PNG
>>
>>109510941
>>
>>109510105
If Comfy had been less arrogant and donated that money to MUGEN, today we'd have local NAI 4.5.
>>
>>109509929
They're retarded. There's a reason everyone switched to DiTs, you want the model to have minimal inductive biases, and you want it to be able to learn to use that extra compute for the same model size that attention gives you.

NAI models are good because they train them for literally 1000 epochs on a 20m+ sized dataset. The NAI4.5 unet architecture isn't good. Anima is mostly competitive with it, despite being trained with a fraction of the budget, and probably on a much smaller/worse dataset.
>>
>>109511164
>They're retarded
They are not. They are a professional anime diffusion corpo, they know what they're doing.
>There's a reason everyone switched to DiTs
>everyone
Who the fuck is "everyone"? The only ones you know that switched to DiT are open weights local models. But you don't know shit about the architecture of closed source models. Or do you actually know what the hell GPT Image, Nano Banana, Midjourney or WAN 3 use under the hood? Didn't think so.
>>
>>109511164
Mfs are larping insider knowledge now.
>>
https://www.cbr.com/sailor-moon-2027-calendar-release/
>International Sailor Moon Collectors Get Official Calendar Release Reveal for 2027
LFTF? (Lora for that feel?)
>>
>>109511651
anime coloring gods are eating good
>>
>>109506800
What's the difference with the first lora 0.1?
>>109506908
Anima Pulse was made by the same creators as Animagine, I think. The dev is from Caligostro or something like that, so maybe he knows his stuff.
>>109507124
Shit mergers remorse
>>
>>109511119
Is that a bodyglove or the tone of her skin. Nice pic by the way.
>>
I dont watch anime, I just goon to anime girls
>>
>>109510432
Nice, NovelAI won majorly
>>
File: 000022.png (2.49 MB, 1664x2432)
2.49 MB PNG
>>109497547
why would you stare at the feet when there's a nice ass right there?
>>
File: ryza final.png (2.51 MB, 1024x2048)
2.51 MB PNG
>>109505465
>be me, a dumbass
>unlock the power of higher res necessary for the model to be able to handle more details
>>
>>109512978
Ultra neat anon, how did/do you do/did (choose whatever i'm esl and don't know the correct order) that?
>>
>>109512834
Didn't realize foot fetish stuff got this normalized. Used to be taboo or not maonstream online pre AI. Now any imageboard has boobs, ass, cowboy shot, **feet**, like it's a standard pose shot.
>>
I request a spoonfeed, what models could I use with just 4 giggers of video memory on my nvidia laptop chip?
Last time I was using SDXL, what about Kria?

I also want to use those C++ projects that make SDXL more efficient.
>>
Any way to tag artists in krea2?
>>
>>109512978
Looks super good. See if you can get more emphasis on her socks squeezing her thighs. Never played the ryza games but that was my favorite part of her design from the box art or whatever. Her thick fuckin anime thighs getting squeezed and you can see it hnng.
>>
julien status?
>>
>>109513362
It just means a 512x1024 image gets fucked up from how stable diffusion models work. I don't know the accurate and technical explanation off of my head, but basically the model gets "confused" and "doesn't know how to map" the details into a small output, hence the low quality. Portraits are good at 1MP, but full body shots likely need to go higher. A small image isn't "a big image but downscaled".
>>109513600
I deliberately put (skindentation:0.8) in negative because >>>/h/8926605 was just unpleasant to me. Her official visuals aren't *that* puffed, and it's not like I removed all of it. I also gave the tiddy a huge buff.
>>
>>109513474
idk but most of the loras are trash or don't even work
Is it hard to make good krea2 loras or what?
>>
>>109513692
Looks super good man
>>
Gow much longer until completely sandbox AI visual novels are here? 1 year? 2?
I mean generated completely from dialogue, images and storyline.
>>
>>109514897
isn't there already ones on cloud?
>>
>>109512592
you should try it
>>
File: 1768872635191263.png (3.54 MB, 2048x1024)
3.54 MB PNG
>cant do goon images
>fucked up hands
NiggerAI4.5 ladies and gentlemen
>>
>>109515417
>>>/h/8926910
>>
>>
>>109513418
Anima
SDXL
>>109513866
Computationally expensive and the model is inexpressive, plus it has some captioning quirks that aren't intuitive. For example, if you caption something as "anime illustration" the model automatically slaps on the slop coloring it always uses for its anime illustrations.
>>109514897
But doesn't NovelAI image gen, when synced through SillyTavern, enables that if you activate VN mode?
>>
>>109515417
>>109515424
I get you anon, but NAI is genuinely our best option if you are a scatfag. Nobody out there is thinking about us, and all the local finetuners keep running away and avoiding our dataset
>>
>>109515417
>cant do goon images
can anyone explain where this shitpost comes from? I understand when people say that 4.5's style is ugly because it generally sorta is, but nsfw-wise it's extremely solid
>>
File: 1766589834493014.png (1.73 MB, 1024x1024)
1.73 MB PNG
>but nsfw-wise it's extremely solid
>>
>>109515494
Idk what people expect prompting anime models without artist tags. New to diffusion perhaps?
Also 4.5 is shit at natural language, you should have subbed when V5 came out if you want comics prompted with NL and all.
>>
File: 1765535913299941.png (3.5 MB, 2048x1024)
3.5 MB PNG
>just add artist tag goy, are you new or something?
>>
>>109515537
You can keep cramming 50 panels in 1 megapixel, type rating:general when you want nsfw, and use default style. You paid for it after all, it's free to use however you like. Just don't cry about shit results?..
>>
File: 1775285324813867.png (2.26 MB, 1536x1024)
2.26 MB PNG
>>109512454
it was meant to be her skin but ai does what it does
>>
>>109515549
stop replying to a bait braindead fucktard
>>
>>109515582
Why?
>>
>>109515549
I'll let you know exactly what's happening shill.My prompt had onee-shota in it so your shitty site censored it. Every shitmix out there can nail it no problem, only your shit site failed

>You paid for it
Used a throwaway email
>>
>>109515593
OK show me a shitmix doing a 1MP gen with 4 coherent panels, onee-shota. It's
>no problem
so it will only take you a minute at most.
>>
>>109515585
shitposter is either ragebaiting you into wasting your time or wants to be spoonfed. either way, why would you encourage such behavior?
>>
>>109515593
anon, while nai4.5 is absolute shit and really poor at NSFW(it will do abortion SD1.5 tier genital anatomy(unless you prompt a very specific short list of ugly as fuck artists(and it still isn't very good even then) and is stylistically dead), the only model you can access "for free" is the curated model, which plainly does not do NSFW.
>>
>>109515450
>But doesn't NovelAI image gen, when synced through SillyTavern, enables that if you activate VN mode?

Setting it up so that the background updates when it should, and the characters designs stay consistent takes a considerably annoying set up, and even then it's very fragile since it relies on many auto scripts and like 3 LLMs working in tandem, and then you are heavily restricted in the way the avatars are shown, you either pre generate all images, then remove the background of all of them, and that defeats the purpose of a sandbox, say you want to introduce a brand new character, you would need an automated pipeline that creates a consistent description for the character and saves it for any future reference, then a diffuser creates visual novel like character portraits with a green/blue background, then a script removes the background and returns it for display, well in the process it often fucks up and returns a wrong composition, for example it may have placed tags describing character Z into the prompt meant for character Y, because currently Z and Y are kissing or something, so you would need another LLM pass to make sure it isn't mixing characters because it read the context wrong or sum.
AND EVEN THEN, it can't do specialized scenes properly, you know in light novels where the generic character portraits are replaced with an actual illustration, well it doesn't transition properly to them, and group scenes are a mess.
>>
>>109515608
No, free trial has full, you just have to switch to it. Also it shows you you need 5 anals for precise reference, yet when you click generate it works.
You can test everything fine if you are not retarded.
>>
File: 1767512008514406.png (1.73 MB, 1024x1024)
1.73 MB PNG
>>109515601
I batch generate hundreds of pics every day using random artists and mixes. Your site is the shittiest
https://files.catbox.moe/b5b5zu.jpg
>>
>>109515552
>but ai does what it does
Yeah. The metal choker makes it look like a semi-cleat body-globe.
>>
How to change character's pose without changing their design? I'm trying controlnet with depth but it didn't really work. Using Anima on ComfyUI
>>
>>109515608
>it will do abortion SD1.5 tier genital anatomy(unless you prompt a very specific short list of ugly as fuck artists
I save some catboxes from various anons to use for my gens. Since you've been at it in 2 different threads now, here are the ones focused on genitalia, in all sorts of styles: https://catbox.moe/c/j0gvbc
Since I save them for meta, they should hopefully all have it. Surely there is enough there for you to find something that isn't "ugly" "abortion"?
https://catbox.moe/c/j0gvbc
>>
File: Krea2_turbo_00055_.png (1.44 MB, 1024x1024)
1.44 MB PNG
>>
Upscalers for Anima?
>>
>>109515436
Nicenicenice
Its the Italian artist, not Vermeer or Rembrandt right?
>>
>>109515480
They tried the trial and didn't switch to the full model lmao
>>
>>109516132
Feel free to generate 1MP image with 4 coherent panels of onee-shota that match my pics >>109515656 shill. One pic is enough, simple backgrounds are fine
>>
>>109516213
V4.5 is not good at comics, I'll admit that. It seems that V5 will be better at it. That doesn't change the fact that your inability to get even a nipple out of the model was simple because you failed to switch to the full model.
>>
>>109514897
You already can do that
>>
>>109516528
Your site used a petty jewish trick to scam new retards into wasting their credits on distilled censored model, that’s not something to be proud of and it’s not my fault. If I had to read posts here just to figure out I needed to switch to the full model to make nsfw how the fuck are other people supposed to find out about it?
>>
File: Krea2_turbo_00019_.jpg (199 KB, 714x1025)
199 KB JPG
>>109516077
Yep, Caravaggio, but i should try also with the Dutch golden age painters, they're cool
>>
File: file.png (2.22 MB, 1120x1440)
2.22 MB PNG
>>
>>109516981
Prompt? Catbox?
>>
update the OP :)
https://animastyles.thetacursed.com/
>>
>>109516987
Here ya go
https://files.catbox.moe/uqwq60.png
>>
File: C168.jpg (358 KB, 896x1152)
358 KB JPG
gaming
>>
File: martyr-chan.jpg (163 KB, 832x1216)
163 KB JPG
>>
I think Anima is amazing
>>
File: tfw no coffee.jpg (154 KB, 832x1216)
154 KB JPG
>>
File: peach surprise.jpg (183 KB, 832x1216)
183 KB JPG
>>
File: ukelala hard gay.jpg (194 KB, 832x1216)
194 KB JPG
cover your eyes bros
>>
File: delf.jpg (184 KB, 832x1216)
184 KB JPG
>>
File: 10000% more gay.jpg (204 KB, 832x1216)
204 KB JPG
>>109517974
here we go
>>
>>
File: foxes.jpg (323 KB, 1024x1024)
323 KB JPG
>>
File: run bitch 2.jpg (314 KB, 1216x832)
314 KB JPG
>>
File: bitch-chan 2.jpg (219 KB, 832x1216)
219 KB JPG
>>
File: sob 3.jpg (261 KB, 1216x832)
261 KB JPG
>>
File: me in the middle.jpg (308 KB, 1216x832)
308 KB JPG
>>
File: things you are aware of.jpg (134 KB, 1216x832)
134 KB JPG
>>
File: hunter bitch-chan.jpg (257 KB, 832x1216)
257 KB JPG
>>109518041
>>
File: planet crusher.jpg (346 KB, 1024x1024)
346 KB JPG
>>
>>109518026
Why did she dress like that if she didn't want trouble in the dark forest?
>>
File: ultimate bitch gal.jpg (192 KB, 832x1216)
192 KB JPG
>>
File: flawless victory.jpg (184 KB, 832x1216)
184 KB JPG
>>109518100
she was just going to the bakery anon
>>
File: lamia-chan.jpg (206 KB, 832x1216)
206 KB JPG
>>
File: hmm, what book is this.jpg (125 KB, 832x1216)
125 KB JPG
1
>>
File: oooo.jpg (122 KB, 832x1216)
122 KB JPG
>>109518140
2. inb4 >where is 3. there is no 3
>>
File: contracts.jpg (125 KB, 832x1216)
125 KB JPG
>>
File: pap.jpg (254 KB, 1024x1024)
254 KB JPG
>>
File: lovey dovey.jpg (339 KB, 1920x1088)
339 KB JPG
>>109518172
>>
File: has-shy-shaku-sama.jpg (149 KB, 832x1216)
149 KB JPG
>>
File: sharky in trouble.jpg (181 KB, 1024x1024)
181 KB JPG
>>
File: fox23.jpg (325 KB, 832x1216)
325 KB JPG
>>109518014
>>
File: harpy-chan 2.jpg (338 KB, 832x1216)
338 KB JPG
>>109518007
>>
File: girl mech.jpg (325 KB, 1024x1024)
325 KB JPG
>>
File: tfw no ubsc charger.jpg (211 KB, 832x1216)
211 KB JPG
>>
File: emma2.jpg (246 KB, 1024x1024)
246 KB JPG
>>
File: www2.jpg (162 KB, 832x1216)
162 KB JPG
>>
File: braap.jpg (255 KB, 1024x1024)
255 KB JPG
ANIS DEPLOYING BIO ATTACK WATCH OUT
>>
File: expression convey.jpg (229 KB, 832x1216)
229 KB JPG
>>
File: tokai teio not a rapist.jpg (243 KB, 832x1216)
243 KB JPG
>>
>>109518218
>>109518306
>>109518318
>>109518328
Catbox
>>
File: freeze.jpg (137 KB, 832x1216)
137 KB JPG
y'all better comply with officer sherry
>>
File: vampire hunter.jpg (177 KB, 1216x832)
177 KB JPG
>>
File: cm.jpg (169 KB, 832x1216)
169 KB JPG
>>
File: baa.jpg (139 KB, 832x1216)
139 KB JPG
>>
File: lusamine.jpg (131 KB, 832x1216)
131 KB JPG
>>
File: oh no a spooky ghost.jpg (181 KB, 1216x832)
181 KB JPG
bros what would u do if a spooky ghost like picrel showed up in your room at night
>>
>>109518338
>>109518381
Catbox
>>
File: 2big.jpg (149 KB, 832x1216)
149 KB JPG
>>
File: rain smoke.jpg (267 KB, 832x1216)
267 KB JPG
>>
File: weanus.jpg (178 KB, 1024x1024)
178 KB JPG
picrel : most disturbing enemies in video games
>>
File: sunscreen.jpg (249 KB, 1024x1024)
249 KB JPG
applying sunscreen is important during summer, fyi
>>
File: sister gyarukina.jpg (145 KB, 832x1216)
145 KB JPG
>>
File: ojou beach.jpg (333 KB, 832x1216)
333 KB JPG
>>
File: delf32.png (1.08 MB, 832x1216)
1.08 MB PNG
>>
File: lewd.jpg (218 KB, 832x1216)
218 KB JPG
>>109518150
more lewd books
>>
File: ara ara intensifies.jpg (219 KB, 832x1216)
219 KB JPG
>>
File: emma.jpg (224 KB, 1024x1024)
224 KB JPG
my job is done. happy slopping bros
>>
File: file.png (2.18 MB, 1120x1440)
2.18 MB PNG
okay
>>
File: 1762654813350953.png (3.75 MB, 1280x1856)
3.75 MB PNG
>>109515739
i wish there was a way to finely customize clothing and accessories, but then idk how that would even work
>>
>>109519090
virtual try-on is what the method is called in corpospeak. you might find something similar, it's pretty common
>>
>>109518847
ogey
>>
krea booru tune when?
anima is just too small
>>
>>109520472
there is a minimax image/edit model cooking so we might just end up skipping
>>
File: eresh.jpg (723 KB, 896x1296)
723 KB JPG
>>109520472
works for my 1girls
>>
Nai 5???
>>
>>109521642
not today, schizo.
>>
>https://civitai.red/models/2849150/carmin-roux-marin-kitagawa-my-dress-up-darling-ai-generated-imageset-style-anima?modelVersionId=3217182
>https://civitai.red/models/2810332/marin-kitagawa-my-dress-up-darling-anima
2 Marin Kitagawa loras today...
>>
>>109521653
Nai v6????
>>
File: 1777771687434349.png (2.35 MB, 1024x1536)
2.35 MB PNG
>>
>>109520472
Way too expensive. Anima cost $50k to $100k to train and that was only 2B params. Now multiply that for a model several times bigger. And people already aren't happy with Krea's loras, outputs are coming out stiff.
>>
>>
>>109521782
I've had some nice outputs with the kroma lora though
>>
>>
this has to be one of the slowest generals ever. Is it always like this?
>>
File: 1762628706078626.jpg (1.25 MB, 1536x1536)
1.25 MB JPG



[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.