Previous: >>110168089No collage edition (don't know how to do it, sowwy)>UIsComfyUI: https://github.com/comfyanonymous/ComfyUISwarmUI: https://github.com/mcmonkeyprojects/SwarmUIKritaAI: https://github.com/Acly/krita-ai-diffusionForge Classic/Neo: https://github.com/Haoming02/sd-webui-forge-classicInvokeAI: https://github.com/invoke-ai/InvokeAI>Image ModelsIL: https://rentry.org/illustrious_checkpointsAnima: https://huggingface.co/circlestone-labs/AnimaKrea: https://huggingface.co/krea/Krea-2-TurboNovelAI (paid): https://novelai.netMoss: https://huggingface.co/Comfy-Org/MOSS-Image>Video ModelsH3: https://huggingface.co/Comfy-Org/MiniMax-H3Turbo lora: included above>How to generate anime:image : https://rentry.org/comfyui_guide_1girlvideo: https://rentry.org/wan22ldgguide (No H3 guide yet)>Style ExplorersAnima: https://anima.mooshieblob.com/ | https://animadex.net/Illustrious: https://tagexplorer.github.io>Checkpoints, LoRAs, Upscalers, & Workflowshttps://civitai.comhttps://civitaiarchive.comhttps://tensor.arthttps://openmodeldb.infohttps://www.seaart.aihttps://www.liblib.art/>Resources:Share Metadata: https://litterbox.catbox.moeImg2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-beta-oneOnline metadata viewer SD/NovelAI: https://spell.novelai.devCatbox/Metadata Userscript: https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fbOutput 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
I too do not know how the collage, unfortunately
>>110190063https://rentry.org/neo_collage
>>110190223Kek I'm getting some interspecies reviewers vibes here
>>110190125ty
what's the point of posting in this fucking general if OP is not gonna take the time to make the collage
what's the point of doing anything when it will be forgotten and erased by time
>>110190332Are you asking that as a philosophical?I liked your gens from last thread
I use NAI but want to switch to local. Being able to strongly copy a style and characters is my priority. I hear Anima just got its style/character reference and is more anime focused but Moss is new and really good. Which way?
>>110190355The discussions in these generals are always centered about dumb retards arguing on which model is good or bad. Nobody comments on anyone else's works unless they look exactly as the source material which I find stupid because anyone with a lora can do that. Only thing anyone who posts gens and doesn't dwelve on garbage discussions can look forward to is getting featured in the collage which is already not a big achievement but it's nice
>>110190382the anima reference seems like its more suited for characters. not sure if it can do style. i guess i should try it out myself since vibe transfer was one of the main reasons i used nai
>>110190395The image says style so I hope so.
>>110190414yeah. what was worrying was that right now character and style are joined but with my vibe transfers i wasnt trying to use the character at all. ill still try it out soon once i stop being lazy
>>110190433True. I think from>Character and style from one imageThe and is the kicker. You can't have one without the other. So I'm thinking that if I go with Anima I'm still going to have to train some Loras.
>>110190386
>>110190453me on the right
>>110190386I prefer engage with gens and catbox directly rather than yapping like a troon.Catbox links are enough on their own so I don't see much point in discussing or questioning things further.
>>110190453Me on the slop
>>110190636Isn't the catbox just an external link to the picture? And regardless I'm just talking about OP being a humongous faggot and not making the collage. Then again OP is always a faggot
anon is always free to make and post their own collage
>>110190660Whoever made this thread did it just to get his own shit going and be the first one. He doesn't give a shit about the posters here so of course he doesn't make the collage, but because his asshole was puckering with excitement about making the thread he rushed it and didn't allow someone who does care to make it. I don't give a shit about me but other anons do like getting featured therePoint being that OP is a faggot, today, tomorrow and whenever
>>110190704nigga the thread was like on page 8. wasnt no one rushin to make the thread lol
>>110190745then let it die, you dumb mongoloid
>>110190761nah cuhcuh, /adt/ is here to stay no matter how much it makes you seethe
>>110190223I was trying to do some convenient censoring the other day and it wasn't going very well. I think IL was way better at it than anima, or maybe it was pony that I'm remembering.
>>110190858IL is quite good with it
>>110191098>>110191038>>110190933>>110190919>>110190871>>110190858>>110190704>>110190660>>110190539>>110190453all of these are good by local standards
>>110191118How do you can do that wothput triggering the spam?
>>110191139I am the novelai supervisor
>>110191151Who? Uragan, Sigurd or Raezulo? Anime screemshot sucks on v5full tho
>>110191165I was joking but who are those
>>110191170Nai employers
Futa?
brats these days man...
had to make another one
>>110191151Hi hello! How would you rate me?https://files.catbox.moe/oz13co.jpg
>>110190704>Whoever made this thread did it just to get his own shit going and be the first one.Thread was on page 9 and nobody else baked.>He doesn't give a shit about the posters hereThat's not inherently something negative.>so of course he doesn't make the collageI don't know how to make one and I'm not going to bother with scripts or anything like that.>but because his asshole was puckering with excitement about making the thread he rushed it and didn't allow someone who does care to make itThread was on page 9, no bakers in sight, there was nobody else waiting to make it.>I don't give a shit about meYes you do, quite a lot actually. You started out posting videos, but once OP began making the collage with only images and no videos, your png persona showed up.>other anons do like getting featured thereYeah, you do.
And if you keep replying to me I'm going to keep increasing the line spacing between each quote.
>>110191404Why she has face of ash kethum , dragon ball, 90s anime?
>>110191427yeah figured I was only talking to faggot OP. I stopped making videos for the time being because they cost money and time to make. I guess your broke ass wouldn't know that so fine I guess
Is there a NovelAI equivalent for video generations yet?
>>110191450Seedance, ViduAI. I've seen people making some cool stuff there
>>110191450Not necessary, H3 is competent, its just very power hungry, but so is nai since a 32 channel vae almost certainly implies 30GB+
>>110191439because she's 90s anime??
>>110191489https://files.catbox.moe/0z6i8v.mp4I'm having fun while screwing up
>>110191736Looks great, anon. My suggestion would be to ask another AI for better prompts so you get stuff like decent physics and the like. When I wrote my own I often used redundant words or repeated too much a characteristic that ended polluting the video or something
>>110191766Thanks! Will do
https://youtu.be/-sWnEWpS_fA?si=iZMtLtOW_W7xv9hQ
>>110191352
>>110191736Looks better at 0.75x
>>110189963real
>>110189939This isn't thread #156, it's somewhere around #165 to #170 (archive link below):https://desuarchive.org/_/search/boards/g.desu.meta/subject/%2Fadt%2F%20-%20Anime%20Diffusion%20Thread%20%23/page/1/Some shizo has been baking without updating the thread number and it looks like a bad faith move to wipe out part of /adt/'s history.
I need money
>>110192160Very cute!>>110192754Very sexy MILF.
>>110191038yes
>>110191225cute!
>>110191433
>>110191736nice, I also like the sounds of it
>>110192926At least post AI generated reaction image.
This thread is 114514 times more faggier than Astolfo
>>110192939ya sowwy
>>110191427>Yes you do, quite a lot actually. You started out posting videos, but once OP began making the collage with only images and no videos, your png persona showed up.HahahahahaSick burn bro. You got him.
New noob DiT +10b model comming out this year.T2i + edit
>>110193364DiT is slow as shit. Imagine waiting 5 times more for a gen than anima.Couldn't be me. I'm not patient enough for Anima already.
>>110193364>DiTDOA architecture for anime.>10bExpensive to train, not VRAMlet friendly.>t2iProbably undertrained because of that.>+ EditNoob's edit dataset is dogshit.
>>110193364>noobWhat does that even mean? I'm new at this diffusion thing but I keep hearing about it.
>>110193364Neat, might finally pressure NAI to release PR and VR sooner.>>110193396It was the first SOTA SDXL anime model back in 2024. Not so much for breaking new ground architecturally, but mainly for how well it focused on styles. The model is unstable as hell and has issues with hands, feet, limbs, etc., but back then it was rivaling NAI v3.
>lax fell for the krea meme
>>110193443Well, at least it's built on a really good base model. I'll actually be able to create characters in non gooning situations with a strong grasp of real life concepts. NAI doesn't understand real world slice of life concepts, which makes it pretty weak for long form, non gooning manga.
>>110193471>non gooning situationsBut why?Isn't it the point to make more goon stuff?
>>110193364Soft reminder Noob is built on top of Illustrious, which is built on top of some other Kohaku finetune, which is built on top of SDXL.Interested nonetheless, but I remember the 3D cosplay dataset
I also remember the failed ControlNet bakes arc with Anima. Everyone acts like Lax made some SDXL breakthrough but he rode on top of work that already existed. He didn't build anything himself, he is a chinese scammer after all.
>>110193493I want to make VNs, manga, and anime. For that I need the model to know any real world thing, in any style I want. NAI is good to Harry Potter vs Malfloid 4 pannels meme.
> i want an entire professional anime production team in one model locally for free no mistakes because... because i feel like it!Yeah, look him in the eyes and tell him that
>>110193621
>>110193364Oekaki bros? Will our mc finally get a job and buy a gpu??
>>110193702no. if i can't run it then oh well
>>110193047The placement of the navel and the abs is not done well
As a Uni student i get free premium access to gemini I was wondering how strict they are, can i use it to caption dataset or use nano banana and other stuff like ERP?
>>110193860the workspace admin can see all your stuff if they want to, btw
>>110193364A real alternative for nai appears.>>110193385You have no idea of anything.
>>110193523LAX and Kohaku are Chinese scammers working for Comfy. Maybe this next failbake makes people stop worshipping LAX and start giving credit to tdrusell whose work on Anima is better and stands on its own.
>DiT is doa for anime despite the only good anime model being DiTCope alert
I don't know what a DiT is or does, sorry.
>>110193994Transformer architecture, nvidia went from CNN to DiT too.
Noob DiT might very well be in the ballpark of chroma, flux or krea in terms of size if its gonna be a 10B model. Vramlets are gonna be uspet, but this is a move forward im happy about. Could be a real nai at home moment.
>>110193985NAI is UNet, end of discussion. >>110194063LAX alone can't do shit.
>>110194125Nai v5 is DiT, stop pretending you're smart.Wait i think this is ani since he repeats that weird spiel.
>not unet>not 32ch VAE>not build from scratch anime focused model>not turbo variantYeah, it's DOA time!
trani studio melty
This ani guy is gonna sabotage noob over his petty wrapper, someone lock him in the squishy room already
>>110194133NAI is a UNet, Kurumuz said so since it's cheaper than DiT. And yes, I'm not smart and I have a lot to learn.
>>110194164Just google it stop being insane you mongolic retard
>>110194177Stop shitposting, NAIv5 is built on UNET is an already known fact.
none of the architecture shit matters if he still thinks anime is only danbooru + e621
>Ani this>Julien that>Local this>NAI that
>>110194177
>>110194192Lax is a lazy chinese scammer painterlyfag faggot
>112 / 36the decline...
>>110194210>nai knows who nozomi is>anima knows who nozomi is>moss knows who nozomi is>krea doesn't know who nozomi isTotal Krea Death
If its a u-net model then noob dit means nothing, you still wont have local peak....
Anima is better in mostly everything else but even in current year I keep coming back to Illu/Noob based models because of how nice the texturing of SDXL is.
>>110194220I for one salute our chinese friends giving us the gift of e621+danbooru
>>110194242I remember some fag posting AI slops of BA girls made in Krea on here and some of them are Moetron-tier abominations
If you see someone hating on laxhar lab or any other competent team pushing out new models, its most definitely ani.
>>110194273>lax> competent teamAny fact supporting this?>it's most definitely Ani.Any fact supporting this?
>>110194289idiot is speaking gibberish
>>110194293H-he made n-noob!
>>110194293They changed the whole scene with noobai XL and now they are gonna do it again. You sit here pretending to be a stranger defending ani.Hope you are banned from all of their socials.
>>110194309Wrong. All he did was put a big strawberry on a cake that was already made. You wont save local with a strawberry.
>>11019432038 star staus?
It sure doesnt mean that it will dethrone nai, but at the same time you dont want mentally unwell schizos even speaking to people responsible for noobai and illustrious.
>>110194311>Illustrious changed the whole sceneFixed that for youThose who were around when Noob are very aware of what's about to come.
>>110194339So? You would be foaming at the mouth if it was an Illustrious model anyway.
>>110194334It will never dethrone NAI in a 1v1, but the majority of NAI users are casuals and if KreaNoob gets placed behind the blue or yellow buzz on CivitAI by demand without needing a subscription, it could pose some real rivalry.
>>110194354No foaming, just setting myself expectations. It could be SOTA, it could be pony tier. Let's see.Can't wait to see what advice the Noob team will receive this time around.
>>110194360Hmmm, at 10B parameters it might just be a bit too much for people to train loras on it with limited memory. But its big enough of a model that it might cover so many topics out of the box that it feels like nai in terms of reach. This is exciting. It could definitely be a nice on demand model for civitai.
>>110194373Alr you seem mentally well, I apologize for calling you ani.
hope it can do some good farting_on_prey. anima is sorely lacking when it comes to this
>>110194397Or good netorare artists, like soraora >>110192754
moss will forever be a 3DPD model
>>110194380>be a bit too much for people to train loras on it with limited memory.people already making LoRAs for Krea 2 since release and with the unlock for Nvidia gpus should make local training even easier too, even for us 12gb vramlets.
>>110194430>+55% speed for sdxlThis is, without a doubt, the pinkest elephant in the room I have seen.
always funny to see a new model announcement and localsissies immediately begin asking how trainable it islike they're so buckbroken they can't imagine in their wildest dreams that a model can simply be good and not need further training
but what if the next new zzz character comes out right after the model release
>>110194518Lmfao. Imagine if models omitted all that slop and only focused on manga/hentai artists and anime screenshots. Would be kino of the highest order.
>>110194540/hdg/ would never recover
>>110194320I could go for a box of stawberries right now
>>110194476Truke
>>110194476 (have a You)Is there a model that exists that is simply good enough to not need further training?Could it be that some people can imagine one but assume that one does not exist yet thus must ask the next most relevant question regarding the relevance of a model?
>>110194600He was referring to nai specifically, yes, that model exists.
>>110194612>no reference
>>110194616Bla bla you dont believe it no one cares
>>110194600I think what he was trying to get at was we locals, are conditioned to assume a local model must be undertrained or incomplete by default.
>>110193621wait is there an african anime studio I can hire to make a series for me? or is every call to an image model actually calls to african sweatshops?
A local model is not born undertrained, it is trained deficient the moment it is named "local", a category that exists in relation to its absent other, the cloud hosted models, against which all local models are measured.
Wow, these threads are moving pretty quickly these days. Glad to see it.
>>110195002/adt/ hasn't grown at all. It's just a group of shitposting hooligans rotating through every diffusion thread.
>>110194211Does Kurumuz have any academic background in AI or diffusion or is he repeating what his employees tell him?
No v5.5....
>>110195082It's inevitable that diffusion threads will attract mentally unwell trolls. That's neither here nor there. NPCs find AI scary and they lash out, it's what they do. Fact is I'm seeing more activity here and cool gens.
>>110192351I would give her a slow, closed eye blowjob, worshiping every inch of her penis with devotion. I would take my time savoring it, sliding my lips and tongue focusing on her pleasure. My hands would caress her thighs, occasionally taking her deeper.
I need more girls tilting back their chairs.
>>110195358Who is this?
>>110195399my llavor of the day gf
>>110195421Who? I've seen her before.
>>110195265Gay in actionsStraight in feelingsVeredict: based
>>110195399>>110195452
>>110195529>>110194071w-would you mind if I... I put my little weenie in your belly button, miss? >///<
>>110195624>>110195265Jack off dude
>>110195569Yeah. Thanks.
>>110195569Mind sharing box for this?
>>110195624I think she would mind, actually
>>110195770Just NB2.1 from Google Flow.
>>110195920ew
>>110195358I would say it made me mistake her legs position from the distance.
I would love to see what the fuck Moss team used to train moss with for anime. Yes I know >undertrained but it seems like they attempted to try conceptualize outfits in some way
>>110196027Don't understand but moss base was heavy gpt slopped in anime to fit benchmarks
>>110190223lora? or model/prompt? please
>>110196037Let me tell you the story about Moss.Act 1: Yay! Let's do another SFW slop realist base model as a university research team, they will give us funds, we can test fancy things like a big text encoder and Flux2VAE! We also have a synthetic dataset with custom concepts to make the model have more prompt adherence.Act 2: Ohhh... I think this realism project isn't going well, it's unstable as hell and I don't know why (probably because of the synthetic dataset), the model doesn't give good real skin and details. :SAct 3: I have an idea! How about we add anime and general illustrations to it and make it like Krea 2? We know that model got quite popular because of famous characters!Act 4: Oh fuck, that didn't go so well, the illustrations and anime weren't learned well, and mixing them with our realism dataset didn't go well either, and it ended up giga slopped.Act 5: Okay, we can save this sinking ship, let's do the Minimax H3 way and add NSFW! With this we can save this model and our failed training attempt!
>>110196053NTA but he uses NAI. He has a lot of gens with similar style you can ask him for so you can make a lora.
>>110196038Signature racing outfit vs signature school outfit vs signature casual outfit (incorrect)It seems like they did something, but it's very vague to know what.
>>110196086woah I didn't know a member of the team lurked here. when is moss-anime coming out?
>>110195168No need to improve on perfection, maybe add a ref2img for gacha sloppers
>>110196086It's absolutely hilarious how much you type
>>110196020That might make more sense idk, looks like lower foot could be closer to left chair leg, or somewhere between where the chair legs would be.
>>110196693He seethes at anything that moves, if he invested that time in fixing his UI, his reality would be different.
>>110196712At least her lower body has a bit more attitude.
>>110195172exquisite robo
>>110195358
>>110196572This looks very nice, the non-body details as well as the body
>>110193364But Tdrusell still hasn't recovered even two thirds of what he invested in Anima, poor guy.
>>110193364Only a retard thinks bigger number = better lol. Can already tell it's DOA like their last attempt post noob.
>>110193364
>>110193364>expected to launchAlso, isn't "expected to launch" the kind of phrase you would only hear in the SaaS world?
>>110197704Nope
>>110197704Yep
>>110197704Maybe
Frankly if novelai isnt DiT, I dont see any reason to not use u-nets.
>>110191225>>110195172OK that's really Cool&Cute.
>>110196053Artist is Fishine, really should be a strong enough tag to stand on its own>>110196126This~
>>110199409Elégance magnifique ! quelle subtilité et finesse !! Bravo, bravo, maître !
>>110199684C'est un véritable festival de raffinement. Bavo, bravo !
I <3 adding pointless text to my gens with the power of Moss
>>110200680I would like to also. But can't think of what text to add.
>>110200680Wait till you try a real model
>>110200997But real model can't do boobs.
>>110201033What?
>>110201057What?
>>110201077In the butt
>>110200997But a real model can't do sweaty umas
>>110201227With big tits and huge horsecocks?
>>110201238It absolutely can
Can we see some gens made on a 'real model'?
>>110201444You mean flux 3?
>>110201444
>>110201477>completely nonsensical rail placement>non euclidean roofs>no boobs
>>110201477Novelai would've accurately modeled the indoors of those building and rendered the skin micropores of those tiny citizens.
>>110201284loras?
>>110201559Lalatina best gurl!
>>110201559Yeah
https://donmai.moe/posts?page=5&tags=ligne_claire
Can local ai do cool spidermen? (He has an anime face under the mask so its on topic)
>>110195172>more activity here and cool gens.I don't buy it. I keep seeing the same burned out avatarfags posting the same gens that define them as avatarfags and people from other anime generals doing charity posting, like you're doing right now It's been over a year as a thread and it still never built its own audience, it runs on anons borrowed from other generals. We all know /ldg/ is the sun that outshines us, there's a lot of topic overlap, and since that thread moves faster and is full of anons who are into the hobby, it's easier to get replies over there.
>>110201766simpler to call it cosplay Crash Bandicoot is a Japanese brand you could use that to troll too
>>110200997But real model cant do anons mom
What is anon preparing for Halloween?
>>110201843I already have the rope!
>>110201847Me too sis, I was planning to do it around Christmas, but then I saw the announcement for Krea Noob.
>>110201893Id be more excited about Noob DiT. Whats krea noob gonna do, a miku doing the robot pose?
I'm creating a plugin that can generate anime from 3d environments, this is the first working attempt, the up video is just a reference, the plugin works on the 3d data itself, it doesn't just apply toon filters to recordings, it generates 24 frames per second within the 3d environment.
>>110201843Making an RTX 5070 costume for my waifu and training a LoRA of it but first I making the dataset with GPT Image (local could never).
>>110201954>anime version>counter strike 1.6 slopLol
>>110201980If you add anistudio to any diffusion op you have my atention
>>110201954it's a start I guess, I think I've seen something similar with shit where some basic geometry shapes animated in blender used as video reference to overlay a scene in motion
>>110201975If you extract frames separately, they are anime drawings. I need to fix the half cut bricks issue though, it's kinda unsightly.Still better than MAD studio animations regardless.
>>110201954Interesting. Are the 3D objects also AI generated, or do I have to make them myself? If I have to make them myself, wouldn't it be easier to use the Assistant Tool in Krita with a Perspective or Vanishing Point assistant and draw them myself, then run them through ControlNet?I'm not mocking your effort, just putting forward other workarounds for handling perspective.
>>110201968Cute
>controlnet>almost 2027Lol
>>110202014Completely AI made but you can give it your own made if you prefer. The plugin has a feature to generate surveyor-like blueprints from which it generates the 3d structures, that's how it made the base model.
>>110201997>>110201850
>>110201954I really need to figure out how to use 3D or pose studio apps to make a base for my images. Describing poses and correct backgrouns are bane of me.
>>110202030If it can generate structures on its own it's a success. Godspeed, anon, good work. Cascadeur does something similar with characters and poses are AI enhanced so they are more fluid.
>>110202056ChatGPT 6.1 Sol Ultra in Work mode can use plugins very well. The way my plugin works is:>give description of environment>it generates the cad file>uses it as guideline in blender to make the 3d>describe the scene you want to shoot in it>...>profit
>>110202063I also made that separately but needs still some work because it looks too much like live2d animations.
>>110201968Cute, a view from behind pwease?
Here is the first character scene generation attempt but you can consider it deprecated, not what I really needed.
>>110201918Why so noisy? Let me gues... local?
>>110202088>>110202139Neat
>>110202143the reason is poorfag AMD GPUor I would be doing higher resolution
>>110202143I can't believe it's the end of 2026 and l*cal models still can't upscale.
>>110202157It works differently from the environment version. We designed the character with turnarounds, accessories, personality, expressions and beta key visuals, he made a sort of mangaka character book for her, which he uses as reference for animations.
>>110202130If you know any site with free gpt or NBP...
>gpt plus >20/month>get a more powerful image model>get more images per day than naiNAIcucks still paying more for less, lmao
>>110202289lovart but it's like one NBP a day
>>110202289>>110195920
>>110202397>>110202365Made it in arena.ai!
>>110201954plugin for whatshadows are bad>>110202056even with 3d or pose studio it's too time and effort consuming
>>110202449Very nice, but... 5 hands?
preview of noob2 stolen from laxcordthoughts? to hype or not to hype? does it beat nai? are we back? nothing ever happen?>>110201893sensenova is a 10b model too [spoiler]maybe that's the base model...[/spoiler]
>>110202539it won't beat nai or anima or moss-anime but should be better than illustrious
>>110202539Prompt? Without it I can't say much, the model may have hallucinated most of it or the style isn't accurate.
>>110202539dogshit
>>110202539I was trying to do something similar with a demon succubus and an angel, but details like outfits and accessories kept getting swappedmaybe someone knows of a good way to prompt for two individual characters that aren't animu or gacha sluts in anima?
>>110202539Looks pretty sloppy but lets wait and see.>>110202665Then it will be utter dogshit like all other local models
>>110202760people get what they pay for. local models are just failed scrap thrown out to the dogs
im going local for the first time, going to try and install comfyui and then anima and moss (are you able to install two at the same time)? my question is is it easy to delete comfyui and the models from your pc? im not good with tech and if i install it wrong or it doesnt work i dont want it permanently taking up storage space
>>110203142larp
>>110203152why would i larp this? i remember why i never come to this site anymore because some idiot is always there to accuse you of being their schizo or larpingnevermind i asked
>>110203142if you use the comfyui portable version you can just delete the entire folder. can have as many models as your storage allows since you switch them with a drop down menu
>>110203142yeah you can have multiple models at the same time, Also deleting is easy, it's not really "installed" in the traditional sense, all the models are download drag and dropped into their respective folders, every one of them has instructions with links in the template page. Everything is contained within a single folder that you can just nuke anytime
>>110203159I think you're just a retard pretending to not know how to use a PC to pwn the nai users, probably even ani, but thats my 2cents.
>>110203172>>110203173thank you for the actual answers>>110203185retard im going local because i use nai and dont want to rely on a service that can go down, why the fuck do you think im trying to sow division. like i said, why does everyone think youre some kind of bogeyman for asking a genuine question. if i was tech savvy i would have installed local models long ago.
>>110203201Assuming you're truthful; normally people start the other way. They use local until they get fed up with the quality and switch to nai. My advice is to not even jump ship, you will spend years learning a craft that doesnt matter.
>>110203214i started trying to get stable diffusion a few years ago but fucked it up and couldnt find the files to delete it so i switched to nai. i think nai does the best quality imagegen but its the impermanence of it i fear. it can go bankrupt or get taken down any time so i want to go local to not have to worry about that.
>>110203234You're gonna be disappointed when you paste one of your nai prompts into anima or moss, but well, good luck.
>>110203279i know but im hoping local will eventually reach nai v5 quality soon, thats all i want. im not genning amazing things i just want to do pinups in cool styles.
>>110203295Then IMO learn neo forge and comfyui right now, maybe try kreamoss and kroma, and wait for better models.
>>110203316thanks. i want to try anima because of the character reference and moss. and just to check, this is entirely run on my own network and for free? nobody is looking in or im going to accidentally buy something? not trying to sow division or act dumb i really am this stupid
>>110203341It's run on your hardware and if you're paranoid just disconnect your pc from wifi
>>110203341Thing is, if you paste one of your NAI prompts into anima, the disappointment you will feel might kill your creative drive forever.
>>110203358cool thanks>>110203396im going to learn how to use it and get the best output from it. i like nai i really do but with all the shit going down about lawsuits for stolen art and backlash rising against ai and data centers i want another method to gen with.
>>110203413theres like 3 models to consider for anime for local, depending on exactly what you want to do. anima knows the most characters and is the only model that knows danbooru artists. it also supports some natural language like saying a character is on the left or right and can do some text but isnt amazing at iton the opposite end you have krea2 (i suggest the kroma 3.1 finetune). it doesn't know many characters but krea2 images have the highest technical level like really coherent backgrounds, few if none anatomy problems etc.in the middle is moss. it knows more characters than krea2 and does natural language and backgrounds better than anima but doesnt know as many characters and still has some incoherency. also doesn't know danbooru artists at all. moss however excels at amateur style images and cosplay and mixing anime characters with realistic scenarios
>>110203341>im going to accidentally buy something?boomer moment
>>110203234Based, NAI bro. >>110203279At the end of the day being local is a mental illness. Sure, the gens are free, but so is the cope.
>>110202539This is measured by money. We will have to see how much he is going to invest and if it’s under 100k that thing is going to be DOA.
>>110203520Cosplay? How do the prompts work for moss in with cosplay? I have it installed just haven't dicked around with it much.
>>110204034just say something like amateur cosplay photography at the start of the prompt. i think it also works better if you use natural language only instead of a tag based prompt when doing a cosplay photo
>>110203234You can make a free account on NAI and type in their UI without a subscription. Why are you yapping, tho, Yappy Potter?
>>110202449>>110201968adorbs
in pantyhose it's either 4 toes or 6 but never 5 with Animawhat model do I use then? I want nice foot focus in legwear :(
>>110204590NAI
>>110202539Someone just ask themIs this just Krea2?If so, slight hype. Someone actually fucking did it.If not, back to Moss
LAX's problem is still its dataset, too many fan art, twitter artist, white background images and sketches, not enough professional studio quality material. I want to be able to tag Kyoto, Mappa, CloverWorks, Ufotable, along with the franchise, and get professional studio quality and not painterly shit.
>>110204907Anime screenshot style is easy make a lora.
>>110204907Just use ChatGPT
>>110205041Are you shitposting? By that logic why bother training simple design characters at all? If "just make a lora bro" is the excuse for skipping styles, it applies just as much here
>>110205078Retard
every now and then the AI throws me a jumpscare
>>110204907Just use the other trillion other sloppa models that do that kind of style you fucking retard
>carjak banned>no bakercatshit was the baker after all?
>>110205950Why the prompt leak to the image tho?
>>110205976huh?
>>110205976it's called a title bwohere check the meta
>>110205990>no text, teXt: a heart held between black and white wings
>>110206008>>no textit means "no artist signatures" in NAI
>>110205339it's over
Its nais :^)
>>110205990>refuses to show original metadataNo-one titles their shit like that, and it's not an LN-like title either.
Localsisters get down, they're attacking us again
>>110206027it's right there bwo
>>110205990Its fucking over for noobai dit... it never even began........
>>110205990Simple and elegant night dress>>110202539Gacha slop nonsense clothing
>>110206210>teXtis that how you do it?
>>110206260i only typed until 0.6::deadnoodles::,the it did on its own
>>110206303*the rest it did on its own
>>110206303You can override this, when it asks about heavy or no prompting enhancements
>>110206303scary
im this >>110203341 anon from earlier, this is my first local gen. obviously looks shit so are there any written guides or youtube tutorials on how you get good at prompting and making good stuff locally? also is it dependent on your pc? i think mine is powerful enough but not the best pc out there so if you get better gens with more power ill have to settle for this.
>deadnoodles
>>110206397we need to know what model youre using since prompting is different depending on the model
>>110206407anima currently but i want to try moss and another anon suggested krea/kroma
>>110206210That's a different pic from >>110205950though, we can probably assume they're in fact the same prompt beside seed, meaning the text is intentional, and we may guess from contextual cues like >>110202539 that it was meant as a comparison to show nai's better text handling (if not for the simpler reason of just feeling like doing it).
>>110206397Your PC decides what models you can use and how long it takes to gen, it doesn't decide how good the images look.
>>110206411for anima i think a lot of the quality will come from the artist tag. theres some general quality tags like score_7 and masterpiece that might help as well. heres a style resource for anima (there was another one but i cant remember the name atm)https://animastyles.thetacursed.com/this is also quite useful as it contains categories of danbooru tagshttps://danbooru.donmai.us/wiki_pages/tag_groups
>>110206433>>110205950 was obviously upscaled from the smaller one. I don't think I could go the opposite way and generate a smaller version of an image that looks that similar by chance to fake a prompt.
>>110206411Anima likes booru tags so you can start there. It can also do natural language, mileage may varyMoss knows booru tags, natural language, and is uncensored, but it's underbaked. Most likely you need a lora/finetune with itKrea is probably the best at natural language but doesn't really know much about booru tags. It's also censored to shit and barely knows any relevant character or styles, pray to the gods that a lora exist and it's not a jeeted lora
did anyone ever figure out what the correct thread number was or should we just change it to 170 or something for the next thread?
>>110206510unimportant for this slide general
>>110206510troll question? if you search "/adt/ - Anime Diffusion Thread" subject line in the archive it had gone from #4 (preceded by few unnumbered) all the way to #156 without obvious issues. No reason the next can't be #157.
>>110206573>>110192546someone was complainin. havent bothered to try and verify
>>110206510Go back two threads and go from there. This isn't hard. backlink the last two threads with the correct number
>>110206495Troon
>>110206397Looks good
last call for someone who knows what the fuck theyre doing to bake. otherwise im making one numbered 157 with no collage
I dont know what Im dong but please just pick a cool picture you like
>>110200368I vote this one
I'm back to these threads y'all
aight fresh bake>>110206813>>110206813>>110206813 if someone does a collage ill delete
>>110206819???
>>110206985i fucked up and forgot the subject. rebaked>>110206984>>110206984>>110206984
I slop-coded a character tag database auditor in pytkinter. It lets me speed through each set of character tags and add/remove whatever ccip grounding got wrong. Often ccip grounding is very good, but sometimes it mixes up characters that to humans are obviously different.I'd package my whole toolchain but the likely issue is most people can't run a decent captioner LLM locally for it to be worth it. Also, Mac-only, because those are the only people who ever pay. Sucks, but time and electricity aren't free.
>>110192546>>110206644Some of those are deleted spam threads. 179 search results minus 21 deleted = 158.158 - 5 pre-#4 = 153.153 + 3 (pretend #1-3) = 156.For example, 3 of 4 #146 were deleted, there's two #145 and they're both deleted, so count one of them and we're at #157. We would need to increment by 1 for any that need to be counted again.#26 unnumbered but linked between 27 and 25; 158#52 was deleted; 159#61 unnumbered but linked between 62 and 60; 160#93 had two threads ongoing, one lasting longer, the shorter one 3 days with 165 posts; 160 unless we want to count these as 1 (159)#94 has two non-deleted but one has only 35 posts; 159 (158)#154 had two threads over 320 posts, one was created 2 days after first one was created and died 2 days after first one died; 159 if both this and #154 count as unique number otherwise (157)So at maximum we're at #159, possibly #158, assuming #4 is just #4.(caught after writing above) #56 has two threads over 320 posts created on same day, one lasting only only 1 day and other 3 days longer. Decrease by 1 if we're not counting them as unique number.Looks to me we can ignore #93 (one thread is half size of normal).The main thing is the relatively recent #154 "dupe" which should be counted separately. My count at that point says >>110206987 is #158OPTIONALLY we can consider the two #56 threads as a single #56 and keep >>110206987 as #157 meaning we wouldn't have to change anything.
all of that for a thread created to spite another kek
>>110207122>Looks to me we can ignore #93 (one thread is half size of normal).Why would you do that you retard
>>110207569There are no threads that go +2 whenever there's a fork made on the same day. Though arguably, if a same day fork began unused but stayed alive long enough and then everyone moves to the fork after the first one dies, then the next could be +2.
>>110207647I'm not following but whatever
>>110207647Are you saying that #93 was a fork?
>>110207686Yes, the smaller one, with not only shorter time span but also fully enclosed within the larger thread's span. Generally a higher number means newer than the lower number rather than "total number of active threads".Also, quit making me waste my time instead of ignoring you.
*To clarify, by "ignore" i meant not counting as multiple, not skipping the number altogether.
