Previous: >>109778389>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-classic>Image ModelsIL: https://rentry.org/illustrious_checkpointsAnima: https://huggingface.co/circlestone-labs/AnimaKrea: https://huggingface.co/krea/Krea-2-TurboNovelAI (paid): https://novelai.net>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>Beware inorganic activityhttps://rentry.org/animanon
>>109800581Can you stop baking early?
Second one to Rock and Stone gets fucked in the ass by Astolfo
>>109800522ani is cleaning the backend and fixing how projects and plugins get loaded and unloaded, pretty impressive work, if ani gets that right it makes the thing easier to add new models and workflows without everything turning into shit
>>109800581Why do you post ani updates then tell yourself to shut up and buy an ad?
>>109800626You're not being slick so blatantly sucking your own cock. Get real.
>>109800626>>109800630>>109800628
thanks for bakering
>>109800664Why are you thanking to yourself?
>>109800669He can't stop gaslighting and thanking himself. Everything he complains about Ani for is just shit he actually does
>>109800626ani is doing lots of hard work behind the scenes so anistudio can keep moving forward. big recognition to ani for his effort
>>109800708>>109800740You're not being stealthy. Why don't you buy an ad and stop shitting up /adt/?
Glad to see that diffusion threads are still full of extreme mental illness.
>>109800740actual kino dev making anistudio more stable while everyone else is yapping,
>>109800776It's sad that you think tanking your own reputation like this is doing you any favors.
give ani recognition
>>109800581>that screenshot
>>109800807Give ani love
>>109800740The only thing moving forward is your descent into absolute schizo.
>>109800803Ani is building a reputation by working hard while you seethe and tank your own reputation
>>109800852Yes, anon. We believe you. You're clearly having some kind of breakdown. It'll be okay. No one can tell what you're doing.
>109800852>109800837>109800807>109800776Your life is so empty you have to praise yourself in the third person. Ani is a ghost you invented because no one else would talk to you.
Why C*tjak is seething?
No one cares about your schizos, post gens
No one cares about your gens, post shizos
>>109800761It's called trolls. We get those because anti-AI types are unwell.
>>109800894Nobody knows. It's been like this for years
>>109801183Small feet
Not enough drill for shoving up your ASS?
>>109801571Which model? Very consistent to be Anima and very creative to be Krea.
>>109801613Are you retarded?It's Nano Banana Pro
>>109801599Lol
>>109801752Madness, the amount of detail and data that this model can generate at only 1 megapixel.
>>109801752Kino. Krea?
>>109801836Nai5
>>109801836Why did you drown your image in oil? It's dripping with too much.
>>109801837Oh shit, it does a damn fine Yoshida. Damn fine.
>>109801882It's fun to get oily style once in a while
>>109801899NAI 5 does it all. Local's only real use case at this point is dodging Anlas costs for upscaling or bg removal. Outside of that, local is dogshit if you're serious about style, quality, texture blah blah
>>109801836How do you make Krea loras with blue buzz?
>>109801933>slopper>serious about artlollmao even
>>109801933Amazing that there are retards out there paying for this shis instead of contributing and supporting open source solutions. Hope the niggers at NAI get another data breach
>>109802422how much money have you donated to tdrussel?
>>109802682where/how can i donate? asking unironically. i want to send him some money
>>109800707Lillie!!!
>>109801933I can believe that and I might get around to trying it at some point. Vibe transfer seems fun>>109801955The only upside to using yellow buzz is if you want R+ rated sample images for the epochs. Or in more than one occasion because I didn't have enough blue and I wanted the lora right now
I want to do a Krea2 lora for My Teen Romantic Comedy SNAFU. Should I include season 1? I feel like it's not drawn as well as 2 & 3 but stylistically, perhaps people want it.
>>109802827Maybe make a separate one?
>>109802783How so? I don't even see the option to use blue buzz. My dataset's SFW too, so I should be able to use it for training. I'll try again, but from what I understood, yellow buzz was the only option available to me by the time I got there.
>>109802422Krea's potential dwarfs even a well trained NAI5. Problem is a real finetune runs about 1M in training alone, team not included. And Circlestone is sitting on a goldmine and doing nothing with it. Anima's got over a million downloads on Hugging Face A L O NE. Charge $1 a download and he'd have paid off the whole thing several times over, with cash to spare for the next model, instead total mismanagement.
>>109803017>Charge $1 a downloadYou don't know what you're talking about. Most of them are from thirdies or dumb SaaS inflating the counter, they won't pay a cent for your shit.
>"hey anon (husband)! why dont you check my chest to find a positive surprise?"why shouldn't have generated those images of my self insert holding hands with my waifu, what do i do now?!
>>109802422Local models are a dead end. Compute hardware is stupid expensive and nobody's fixing that, and I'm done pretending otherwise. I'm on uncensored cloud now and not looking back. I don't care about your open weights crusade, I just want to goon.
>>109802698not sure since hes with circlestone labs. they might have some sort of donate link somewhere
>>109803214>self insertHeh, that's what we call a cuck in denial round these parts, homie
>>109802698Cool, another aesthetic finetune or turbo model gets your support. Anyone with two brain cells who actually uses this stuff would be donating to Xinsir instead.
>>109803269i like the anima finetunes he does
>>109802698Why not support an alternative not python based UI?
>>109803356we need something fast lightweight and versatile and ani is probably the only one who understands what that means
>>109801183soup
>>109796946https://files.catbox.moe/6brkdz.rar@orristerstyleAnimahttps://files.catbox.moe/j63ktu.pngTrained on stuff only made in the past yearhttps://x.com/orristerioso/
>>109803574Lol
>>109803214Omg, so happy^^
why are the honkers so fucking fat
>>109802422And why would I give money to narcissistic, personal clout chasing lone wolves who hoard datasets, training settings, and dump money into their own DOA artistically opinionated half baked projects instead?No dude, fuck off with your distorted view of things. I'll give my money to whoever deserves it, to the company that invested, bet on themselves, hired AI professionals, and worked for it, not your Discord peers who don't know how the world works and want all the money in the world for themselves to troon around and experiment as many times as they need to.
>>109800581>Generate a lot of images to get v5 limit down before subscription refreshes>Subscription refreshes>Turns out the NAiv5 generation limit does NOT refreshwhat the actual shit
>>109804308Yep, it's snake oily
what's a recommended model currently to edit anime images anon, and I don't mean inpainting, actual edit prompts
>>109804762probably Krea2 + edit lora or klein
I can't drag and drop images for workflows on Comfy now. Is there a way to do that is custom?
>>109804806krea 2 does anime? and is it even an edit model?
>>109803226Hardware for local image gen is super accessible.
>>109804821It does official anime better than anime. no booru or nsfw without loras however
>>109804837>anime*anima*
>>109804020Because it's open source, you run it on your computer, you own it and you fund the project to use it later as much as you like, meaking it cheaper and better for everyone in the long run. That's what big tech companies do by the way and how they build their underlying stacks. They're open source.Now, you're stuck with a closed source shit from a company that can charge whatever they want and is now squeezing you bunch of monkeys for every fucking penny. Good job retards
>>109804837thanks anon, what edit lora do you recommend?
hi
>>109803356>>109803380You're not being stealthy. Buy an ad.
>>109805026Anons only love seeing the lolcow (You) produce milk
>>109802422I donated quite a fair amount to local projects, including dead end crap like jordach and chroma. local is just dead now sadly, it’s way too expensive to finetune which is why all the local models have been 2b shitbakes or 256x256 joke projects. I’d rather just pay the $25 for a month, get a good goon out of it, and then move on for 3 months until I get the urge again
>paying to goonlmfao
>>109804699>>109804772Cute~
>>109804762>>109804806Krea edit lora murders style of the character all the loras all 3dpg pilled better is H3 or Anima
>>109804892Hi princes ฅ(ᵔ꒳ ᵔマ.ᐟ
>>109805768h3 captures the style better?
>>109805878Better yet, h3 do not change the style and the segmentarion is better.
>>109805929how do you use it for image edit only? it's a video model
>>109804853>t. third worlder with 2012 Acer Inspiroi laptop 4gb vram funding people with an RTX 6000 Pro who live in the US or Europe, so upper class hobbyists with 12, 24,32 ,48gb vram get to play for free.
>>109805935Not trained for that so results are hit or miss like usual, just gacha roll seeds till one lands. Away from my PC atm, give me couple hours and I'll post the ComfyUI workflow.
Why don't we ask MrBeast to fund a 1M dollar anime model?
>>109805975thanks anon
>>109805468>>109805480>>109805489Cute hairpin tho
>>109805946What in the world salad are you trying to say you inbred retard?Even a thirdie that can't run it local would benefit from open source, since more providers would be able to host the solutions just like it happens with OpenRouter, and thus making it far cheaper than having a single closed source provider.
>>109802783Cute setting is Cute.
>>109804020I don't put my work on civit.ai for reasons similar to yours - I see no reason to prop up their bullshit buzz "ecosystem" when huggingface just hosts it for free and doesn't seem to care if it's NSFW.I made the dataset to my last project public. It gets around 10K downloads per month, but I think a lot of that is just "download everything" bots. I have an eth wallet address for donations - zero, as expected.SOTA model training was never in the hands of amateurs and single-gpu people. What people want (coherency, text, broad knowledge without loras, prompt comprehension, etc...) just continues to pull away from what even a midsize company can afford. Compounded with hardware hyperinflation, the outlook for local really isn't great at the moment. It also doesn't help they're pushing safetycucking into local hard too.
>>109803574thanks senpai
>>109806328Fat Dog Mendoza
>>109806021good idea
>>109806328Orristeroso completely defeated and sullied. Their artistic style that took years of work, blood sweat and tears, to develop, cloned and copied by ai and replicated effortlessly by coomers on 4chan. Ntr mindbreak doujinshi starring Orristeroso, Orristeroso violated by burly buff chad AI androids.
Anime edit model please...
>>109806673compared to anima stealing from all artists on gelbooru simultaneously adding just one more doesn't seem so bad
>>109806673nothing of value has been lost
>>109806110>imagine not supporting this>imagine not supporting grown, selfless people who understand how the world works and come together to make the greatest SOTA anime model in history, offered as a service for other grown ups who understand how the world works can use it by working and paying for itLmao, I'm not supporting:YouYour favorite state funded lab that keeps making out more censored, lower quality base modelsYour furry finetunerYour lonely, attention starved, misunderstood genius noobcorder with no money either.
>>109806673think of it another way, when the artists finally chokes their style will live on
>>109806843It's like an afternoon project for vcg lmao
Did this for a friend. Hopefully you guys like it too.
>>109806869Do you stroke him too?
>>109806673Orristeroso here, thanks for making a LoRA about me. It will help me make my illustrations faster. Unlike you, I already have my clientele, so this comes in very handy.
>>109806904Nah, we play poker.
>>109806935*penis poker
>>109806931>d/m/yHeh
>>109806945**poker penisSucking cock rn, sorry for the typos
>>109803999uncensored plz
>>109806869Seems like you two don't get along very well..
>>109806855The GUI yes
>>109806977>footfagsorry sir, it wasn't made with your mental illness in mind
>>109806977how many more VAEs until we have good feet?
>>109807002about three
>>109806931I love your art so much man. You draw the cutest softest girls
>>109805489>4kPlacebo or objective observable results? I don't want to waste my precious v5 stamina testing this, there were other anons that wereusing "res multi:1.5x"and so on
do you ever think how some characters are very similar
Well, I’m training a Krea LoRA on CivitAI with Blue Buzz, following the advice of the wise cui_. My question is: how long does 3,000 steps usually take?
>>109807452>3,000 stepsHow many pics in your dataset? you set the steps based on your dataset not some random guide on internet
>>109807476I wanted to overfit so I wouldn't have to train again, and also just to see how far I could push it. 450 pics of the same OC made with NAI, with 8 different outfits. The outfits are secondary, though, I mainly want to see how well it can represent the character face and body. With Anima, I couldn't get a 1:1 representation but a cheap generalization.
reminder: she wonned over any anime girl.
>>109807417This is a hilarious example
>>109807696Girl failures > greedy schemers
>>109806977
https://civitai.red/models/2918004/magetrail-mageflow-4b-dannborue621-finetuneHOLY FUCK MAGE TRAIL UPDATE v0.2!
>>109808067huuuuh
>>109808067This is nuts, the model is learing better than Anima
This is nuts, they're shilling harder than Anima
>>109808128v0.1 is more cute
>>109808144crazy right? a whole 1 post and 3 reactions, I wonder how much they paid bot farms for that insane relentless spam
>>109808128Another comparison, much more coherent than v0.1This model has potential.
>>109808067>Yeah it's only been 6 days, crazy, but thanks to generous support from some donors as well as another round of Banodoco grant (this time 140+ dollars) I've actually managed to gather enough funding for a 170 epoch continuation run to push the model to 200 epoch total (8~ million samples seen). Neat
Anicord model?
>>109808128>learning better than animaHow do you know this? What if anima was learning 10x faster than this during its base model training?
>>109808247Ani's model, please understand sirs
Why would he even bother throwing money at it like that? With $300 or even $700, he is not goin to accomplish shit anyway lmao.
>>109807452Epoch 8/10, and it’s still undertrained. I also trained an Illustrious LoRA for the same number of epochs to compare, and it started learning as early as epoch 2 LOL.Maybe it’s because I included Danbooru tags in the training data?Can I continue training this LoRA on CivitAI from that epoch? I still have 1.5k blue buzz left.
>>109808178v2 looks beter here
>>109808067>* Being 15-20% faster than NVIDIA Cosmos2/Anima on inference despite being 2 billion parameters largerit's slower for me
don't care still using anima
https://civitai.red/models/2935761/anima-control-net-lite-nowa-lineart-trained
>>109808376Unload anima first falseflag
>>109808067Based>>109808391For how long?
>>109800581>he used them as tools min-maxing themUh, yes, that's both his job and ability. I don't see *that* part being the big deal.
>>109808394>tongue on the wrong side on the majority of theseLiterally unusable.
>>109808546Fuck, meant for >>109802065.
>>109808546???
>using a diversity maximized condensed 41k images dataset as a way to tune booru concept and tags based prompting + Illustration capabilities into the model without having to tune with the full booru dataset. (which would have cost 50k+ dollars, no thanks).into the trash it goes
>>109807452>>109807508>>109808341Okay, glad you figured out the blue buzz thing. Number of steps is irrelevant in and of itself without knowing how many images in the dataset and the batch size, in other words how many times each image gets "seen", which civ tells you. I typically shoot for each image getting seen ~30-40 times. Also relevant is the learning rate -- for Krea the standard 0.0001 tends to works fine imo. >450 imagesSo yeah that's going to be undertrained with only 3000 steps I'm afraid. I also go for more checkpoints/epochs.>Continue trainingHmm that might be risky. Might not get you there if your settings are a bit too weak, possibly better to just do it again with more aggressive settings. What is your learning rate and DIM/alpha?
>>109808713The CivitAI recommended settings. Weird thing is, the demo samples on Civit looked utterly shit, but now I'm testing it in Krea Turbo and it looks surprisingly good. It also understands both prose and tags, so I'm very happy with it right now. It worked way better than Anima. I'll keep using it for a while and form an opinion from there.
>>109808676Give it time anonnie
>>109808831Oh yes, forget about the previews, they're useless. Especially because you probably didn't append the trigger(s) to the beginning of the sample prompts, because it stupidly doesn't do that for you. I just brute-force test epochs 12-20 or so. It's time consuming but worthwhileThe default settings are actually fine for Krea (not for Anima) so that shouldn't cause any problemsAnyway glad it's working out
>>109808676why are localkeks like this??>training krea 2?? NO THANKS, time to blast dinglediffusion0.6b until it hallucinates 3-fingered blobs with anime eyes!training these shitbake models isn't a shortcut, you're paying for the foundational training they all lack. 5 epochs on krea is equal to 50 epochs on dogshitterflow.
>>109808885DON NOT CALL ME ANONNIE
>>109808341Testing the Illustrious LoRA alongside the Krea LoRA and my previous Anima LoRA, Illustrious and Anima have way more in common when it comes to interpreting the dataset, while Krea 2 captures the style much more faithfully.Maybe it’s because of the extensive anime background in both Illu and Anima, which makes them interpret the dataset differently? Who knows.
>>109809091or it's because anima and illustrious are 3b shitbakes while krea has 12b parameters worth of knowledge. it's like asking when some 5b llm can't learn to code as well as a 5T GPT model. local is being held back by vramlets and poorfag micromodels
>>109809124And money anon, it's not the same thing risking 20k-50k, which is already a shitload of money, vs risking 500k-1M on a 12b model. Unironically only Unironically, only MrBeast can save us.
>>109808391ani was right about anima being shit tho
>>109809244We don't care about your opinion, Ani. Stop patting your own back.
>>109809244https://vocaroo.com/1mY38G4OvhKp
>>109809244>>109809267>>109809299Please just fuck off CJ
>>109809244anima is dead weight dragging local down, embarrassing honestly. ani would fix this in a weekend if ani felt like it
>>109809299that's surprisingly catchy, i like the flow>>109809329you first, ani :)
>>109809342>>109809329
>>109809342Cringe. You think blatantly sucking your own cock like this is going to make you popular?
>>109809346>>109809350>>109809329
>>109809329Hey I'm just giving a proper ED to the Ani anime>>109809346I think the somg sent him over the edge, mentions of his lust send him in a frenzy
>>109809356>>109809329
>>109809363Are you so fragile that a parody song has you so upset you're crashing out and pretending one anon is roasting you?
>>109809342ani doesnt even need to try, everyone knows it.
>>109809380>Are you so fragileyou already know the answer lol
>>109809380ani already codes better than half this thread
>>109809386model is already worth it's weight in gold, I might have to do a bad apple cover and post the lyrics in the thread for critical damage>>109809389The song has that covered
>>109809389not well enough to have a job LMAO
>>109809389It's why I don't get the hate
>>109809389>>109809406Yes, yes, you're being very subtle. Retard.
>>109806015Revised again because I fucked up the tag
>>109809416His pride is hurt and now he's crashing out.
>>109809244ani was the only one brave enough to point out anima was shit
This is either an outright samefagged argument or one Ani-impersonating troll taking advantage of one disgruntled retard
>>109809494>advantage of one disgruntled retardWho?
>>109809494no anon, ani is actually that retardedjust check his posts that were deleted when he got banned
>>109809494If only we didn't have extensive documentation of him doing this for over a year
The hit dog is hollering
>>109807721Outfeet
>>109809510"Him" being the Ani-impersonating troll that keeps you on a cuck leash?
>>109809547Ah, yes. The good old "everything bad I ever did was someone impersonating me" argument. Like getting booted from Comfy Org for your behavior, that was someone impersonating you as well, am I right?
>>109809547It is getting funny at this point. Catjack is a complete lolcow
>>109809554Let's leave him alone, he got buckbroken over a song. That's punishment enough and I suspect him and his little buddies are going to take it out on both threads now instead of showing wit and making something funny.
I'm tired of this shit man, I going to bake an /adt/ branch on /c/ if anyone wants to talk there is welcome
>>109809568>Let's leave him alone, he got buckbroken over a songWeird, he was the one that generated it
>>109809577Wishing you a speedy recovery
>>109809554The problem is Ani's stalkers are so insane, impersonating him would be one of the milder things they do. Often they're not even subtle about it. One post either earlier this thread or last thread was pretty much "Stop making fun of me-- er, Ani tee hee".
>>109809570is the baker here? can you share the collage script?
>>109809590It's in /ldg/ and the baker is the anon that is sperging just fyi
>>109809587Ani doesn't need any help to make himself look deranged. There's a reason he was booted from the Comfy offices.
>>109809613It's not about help, it's about how insane his stalkers are. Like I said, they're not subtle about it.
>>109809631I'm glad you finally realized multiple anons are tired of your shit ani, it only took a few years
>>109809631fascinating how you're so focused on ani's """stalkers""" instead of the guy that spams threads with self-promotion and has repeatedly splitbaked our slow general "anon"
>>>/c/4473835>>>/c/4473835>>>/c/4473835/c/ branch of /adt/
>>109809679>renty in OPrip
>>109809679Bro. That collage is so fucking out of place in /c/. Realism, /aco/, full nudity. Did you even look at the board before baking there?
>>109809679fuck off and kill yourself you literal retard.if I wanted to resurrect kdg I would've already done it by now and it wouldn't be halfassed like that shit ass garbage.
>>109809781>kdg??
>>109809789>>>/c/4384488
>>109809798/kdg/ - Kawaii Diffusion General Anonymousinteresting, and why it only lasted one thread? not enought people?
>>109809830...do you not notice that there's a backlink in the OP? It lasted several. It just died because it died.
>>109809841So is this for the dev's sake?He will still spam there because he can't help himself
>>109809841holy shit a thread lasting 2 months, madness
/c/adt it's our little little sister
>>109809877well maybe a little of schizo energy to that board will be good
>>109809789kunny diffusion thread?
why are there no fucking images
>>109809968we are being raided by /ldg/
>>109809977>/ldg/ is one obsessed schizo
>>109809968
>>109809977Proof?It's odd how you always make these crazy claims but can never provide evidence yet everyone you fight to defend you ignore the verified evidence.
>>109809998why are you reposting from /c/? >>/c/4473845
>>109806843>selfless peopleLol, you're a complete and absolute retard getting squeezed for every penny.Of course most open source is random junk. People focus their resources on the good projects that have made achivements and have leaderships, with even enterprise plans if needed, that's how we got tools like ComfyUI or Invoke. Let alone other open source giants like Blender who competed with cutting edge tools like Maya. Fuckimg FFmpeg that Youtube and Netflix are built on is a free open source tool.THAT's how the world and technology works. YOU are CATTLE, literally hindering progress for everyone else.
>>109810102>>>/c/4473845
>>109809679are you noodlekek?
>>109808394anyone try this? workflow anywhere? i can't see that image on civitai either
>>109808067that looks pretty cool, i still like anima more atm but i hope it works out. definitely looks promising
>>109811954>>109811961>>109811975>>109811984Nicr
>>109811038It's still a 4B model though, anon. That's just more low parameter slop. 2B, 3B, 4B... these are way too few parameters to be anything more than barely functional.
>>109810104I play soccer with friends every week. I don't feel some moral duty to go learn how soccer balls are made or invest in the local field rental biz. I pay to use the field, I play, I go home, nobody calls me cattle for that.
gotta stop feedin the trolls famalam
>>109811184cute meryl
>>109800581my wife!
>>109812359No one tells you to learn anything.You pay to use the field because you don't own the land, and you are subjected to the landlord rules and whims. Open sourcing is owning the land and raising cows in it if you want.You're putting NAI in a position to become as predatory as Adobe, with the exception that you can't run their products even locally.>>109812417Kys retard
i love Yunyun
sex
>>109813142
>>109809679Very basedThanks for the bake with rentries, it's important
Any local news?
>>109800740this but unironically.
>>109813142>filename Kek
>>109813707is she going to turn into the creature on the sign?
>>109813707>>109814456is this a creppy pasta anomaly or some kind of modern horror thing?
>>109813066>yeah dude, I have the .safetensors file and the .toml, I understand the technology, now I need funds and I blame SaaS because they have money and I don't, they are predators!Owning the land doesn't make you a farmer but before continuing this stupid conversation:-Do you know any local finetuners who have +48gb gpus (like the /ldg/ guy)and may share it for free for his beloved local anime community?-Do you know any local finetuners who have an updated, well captioned, good quality anime dataset current to 2026? -Would they share it with their beloved local community for the sake of training the next good anime model?-Do you know any local finetuners who are up to date on how models are actually finetuned today, rather than just relying on trial and error leftover from the SDXL era? -If so, would they share that knowledge?-Do you know any local finetuners who know how to finetune Krea?-Do you know any local finetuners who know how to do a distilled CFG model and would share that knowledge (for example, the RDBT guy)? This is necessary for Krea since its base model is bare bones. -Did the RDBT guy shared his knoweldge to his beloved local community?-Do you know who finetuned Anima without catastrophic forgetting? No, tdrussell doesn't know how to finetune Anima, that's exactly why he decided not to update the dataset in small incremental updates (like most finetuners wanted to do), and instead chose to stick with his year old dataset and invest his time in turbo LoRAs or aesthetic merges.Taking into account all those things mentioned, is NovelAI really a predator?
>>109800581Any point in using Anima-2.9B or 3.8B compared to anima-base? Is that shit snake oil placebo or is there actual improvement to be seen?Alternatively, are there any finetunes that are actually better than anima-base?
>>1098147252.9B knows more characters. Otherwise it's not better. Everything else is snakeoil.
>decide to eat at home once instead of going out>get access to the sota anime model for a whole month
>>109813362i want her to work on my car
>>109801571that's very nice
>>109814725anima is still shitmore news at 11
>>109800626Actually he still fails at github actions lmaoLolcow "dev" indeed
post your automatic negative prompt me: non whites
>>109816161I use anima turbo, negatives are a thing of the past for me.
Krea will always be a dead end until someone unlocks the ability to goonslop with it.
we moved to /c/ sowwy old (hag) sister :(
every time I think about trying this I just stare at the screen without any creative ideas.maybe I just don't get it? What's the point it's faster to just google an image I'm looking for than try to roll the dice and hope it appears in a thousand gens.say I get that ideal image of a cute girl showing me her ass, then what? look it at a few times then bank it in a folder I'll never touch again. Then repeat the cycle.
>>109816241if all your ideas were already done a thousand times by other people then genning isn't for yousorry, I don't make the rules
>>109816216thicc archive
>>>/c/4473948/g/adt our reponse???
>>109816216some uncanny vibes here
i've had good experiences with anima
>>109816419My main problem is loosing the random machine as a prompt will produce very similar results regardless of the seed.
>>109808713>6 fingers
>>109816341Catjack won, divide and conquer, first ani vs non ani, now /g/ vs /c/
>>>/c/4473957gee, I wonder.
new to genning and trying out illustrious after starting with anima. is it normal to need to detail the fuck out of everything for it to not look wonky, especially with the faces?other people's workflows i've seen looked fairly straight forward and didn't need to do all this bullshit.
>>109816849yes, that's normal
>>109816849While it is normal, you don't need a huge workflow to do it. You only need a single inpainting station. Re-use it for every detailing job you need to do on your image. (Detailing is just high-res inpainting.)
>>109816313
/adt/ tho
Next thread we continue it in /c/
>>109816216>>109816734>>109817031Prompt? Catbox?
>>109817137we need one more general in /jp/
>>109816802But /adt/ locals aren't schizos themselves, those are c*tjak and his /ldg/ thread shitter troupe.
New thread:>>109817250>>109817250>>109817250
What happened to Silly Tavern? Why literally a troon is the main dev? Where is Cohee?
>>109817243and this is their most active congregation point, making it more schizo than their original home threads by default.And this thread is, at this point, legitimately, by all marks, really, really bad.
>>109817263>trans people are succeeding where I'm failingBros...
I'm working on a Yahari Ore no Seishun Love Come wa Machigatteiru dataset, which is about 3500 stills. They've gone through a general cleanup process, removing intro and outtro shots, anything lacking a face, anything without a main character out of my roster of 12. I'm using minimax-m3, which seems to be capable of handling a list of character descriptions for 12 accurately, and otherwise saying "unknown" if it can't find a match. Sure would be nice if insightface worked with anime faces, but it does not. If it did, you could tag images comparing them to reference images with pretty much absolute certainty.
>>109817157pRoMpT? cAtBoX?
>>109816876>>109816987danke
>>109812138i feel like with better algorithms you can still get much better results. anima is smaller and still better than the stuff that came before it. i wish someone would target 12 gb or 16 gb vram but it is what it is
>>109814593A company will be as predatory as you allow it to be. NAI5 and their new policy is a crystal clear example of that.Again, you're confusing lone wolf finetuners that lack the means with actual labs that have made insane open source progress. NAI took that progress, built upon it and made it closed source.Unlike frontier LLM labs that need data centers to run, NAI not only took advantage of open source, but doesn't add any real value on its own that could't have been made by these same open source solutions that run on local hardware.Do what you want, just know that you're setting yourself up.
>>109816849Use Invoke if you want as an alternative, it's better for inpainting and has a non destructive workflow compared to Comfy
