Previous: >>109817250Please stay on topic>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
Encountering strange laws.
>>109848095Law abiding elf.
>>109848095>>109848098based
>>109848129The cutest squid has arrived!
Really telling
>>109848357Someone is not happy about something he posted being bought up. I think we should stop talking about it for the sake of this general.
>https://huggingface.co/lodestones/Kroma/blob/main/kroma-sensei-booru-e6-teacher-velocity-v0.1.safetensorsapparently, it’s a teacher model for OPD distillation. I have no idea how useful it is standalone, but maybe—just maybe—someone could bake with this and we might get a good next gen anime model in Krea.The ball is in your court {{user}}.
>>109848391I barely have any hopes of anything of note happening with lode. The only good model he made that held up for a few months was fluffyrock
>>109848391It's very sloppy from my tests but if you think it should be there I can add it and anons can draw their own conclusions
>>109848492I like this booba shape. Artist tag?
>>109848500> 0.51::hekoningyou (waraningyou)::, 0.7::massakasama::, 0.8::jidao huashi::
>>109841637Omni Flash is too retarded to understand she never turns around.
>>1098484926 7
>>109848555This, on the other hand...I wish I was a professional who knew how to use Blender and shit.
>>109848482Add it where? Nah man, do not add it to the OP, was more of a speculating post
>>109848555>>109848572Which model is omni flash tho?
1-Tho2-Thoebeit3-ThoughtESL here, how can i use this 3 words in coherent form?
>>109848668Forgot, but It must be in the same sentence and the three together
>>109848391>maybe—just maybeFuck , right in the nostalgia, when claude was kino for ERP
>>109848534Thanks bro. Going to take the tags for a spin after work
>>109848656filename
>>109848703Neat>payinghuh, so seedance is worth paying for? have you tried h3 on comfy cloud?
>>109848810You are supposed to email his company for a licence
>Maintenance has started. We are doing necessary preparations to switch the image models over to a separate class of GPU.AAAAIIIIEEE my seed
>>109848814NTA, but Seedance 4.5 is really good, I use it via nanogpt sometimes.
>>109848836Cope and seethe SaaS cuck
>>109848814Comfy cloud prices are really bad
>>109848814Tried expanding my dataset with H3 and shat the bed. References are good, but if you feed it a bunch it uses them as checkpoints to continue the video. Whatever happens between them, the face goes off style.
>>109848904Just buy a licence and make a game you can sell on steam
>>109848814Well I wouldn't say it's particularly good. I was messing with it on rare occasions, costed 29.7 cents for this 7 second clip at this size, and the others are more filter cucked (I don't mean this scene). Omni is just something on Google Flow you can make a few videos a day as free tier.I'm not subscribed to anything since I'm being a miserly retard, rarely into NAI, I'll wait until PR comes to V5.
>>109848957no shit
>>109849286kek, good gen
>>109849286lol
>>109848057my default negative prompt: (non whites:100), (minorities:100), (jews:100)what else should i add?
>>109849286>not using one of those insecticide cans with a tiny pipe at the front
>>109849384Don't reply to the ban evading schizo please! Thanks You!
>>109849384I will consider that next time, I typically use a shoe
>>109849393I didn't realize that was one of the schizos, I don't come here often. Might be best to leave again, too many schizos.
>>109849381I want to see what it looks like in the positives
>>109849438It's literally an image of ranfaggot strawmanning his boogeyman. Sorry in advance if you get clapped in the crossfire
Just rename the thread to /1v1/, I'm out.
haven't genned in like two months. anything new for meta?
>>109849690theres minimax h3 for videos which came out last month or sofor images its still anima and krea2
>>109849009nice style>>109849637Rats are cute :^(
>>109849806>>109849393
>>109849806I didn't have her hurt it for a reason, I was just going with the vermin theme. I'll do more bugs moving forward.
>>109849817no problem ^^
>>109849690Not trying to troll, but if you get the opportunity, try NovelAI V5. If you like anime, it's the best anime model out there. If you come from /aicg/ and places like that, you can find a lot of Chinese proxies hosting it.
>>109849846Sounds like an easy way to infected
>>109849864Nah, I used it years ago, even before I started using local. I got into local through NovelAI after my proxy owner ran out of keys. It’s pretty impersonal, you get the proxy URL, a personal API key, and around 30 50 images per day. If you are in the chinese proxy scene, you can join multiple proxies, so obviously your rate limit goes up.
how do I tag this typical face from frieren?
>>109850185Use emojis/japmojis
>>109850185I see a mix of =_= and :3
It's kind of sad when you realize despite how hard he tries he can't get the same reactions or positive engagement as the other guy.
>mfw I can't goon because I'm sick with the flu and fever. Feels bad, man.
>>109848057If the question is Yunyun... my answer will always be protect, take care, love a lot but also sex
>>109851174Based
>>109850185Looks like Anima plus a Frieren lora to me.
>>109851264maybe the way to make that Frieren face is to train a lora on it.
>>109848057I'm following the wan22ldgguide guide. Where do I put these?>/ldg/ Comfy I2V workflow (Kijai): ldg_2_2_i2v.json (updated 17th November 2025)>/ldg/ Comfy T2V workflow (bullerwins): wan2_2_14B_t2v_example.png (updated 2nd August 2025)
>>109851346.json drag it into ComfyUI main window to load the workflow..png, same thing, both will load the workflow.Let us know when you load it so we can tell you which folder it goes in, anon ^^
Coming soon
>>109851346Just use h3 anon, it's newer and works fine on low vram. wan is an older architecture, and its prompt adherence and wait times are worse. Even the 8 step turbo loras are stable on h3.
>>109851483I think I fucked up?I messed up and put the wan_autoinstall batch file on the wrong folder and I ran it. Realizing my mistake I deleted the files. I moved the batch file in the proper location and tried running it again.These folders aren't being created anymore and I keep getting warnings. I tried ignoring and running the run_nvidia_gpu_optimizations but it doesn't work.
>>109851536Nevermind. I found out why it wasn't working. I had [ and ] in my folder names. I forgot some programs hate that shit.
>>109851785heh
>>109851483Nice, I'm really happy with my Krea lora. One thing I've noticed is that the VAE matters a lot for nailing fine details, but comparing it to Anima, parameter count matters just as much. Both Anima and Krea use the Qwen VAE, yet Krea gives me clean pupils, hands and feet.
>>109851785Lmao
>>109851996Glad you're enjoying the lora bro
>>109852378Haven't seen her anime. Does the background match the show's art style, or is that just the Krea's take?
>>109852512The lora is trained mostly on in-game card art. Idolmaster is first and foremost a game franchise. The anime are not that high budget no. In terms of Cinderella Girls, which is the branch Kirari is in, you've got the 2015 anime which is a lighthearted drama, and then you've got Gekijou which is a bunch of short-form skits based on the comics
>>109852512Any generalist non anime base model has this issue, Krea included I guess.
Training artist style lora for Anima225 pics - 3000 steps - 1.0 learning rate - Prodigy - 1024x1280 Is that ok.
I'm still learning but up until now I've never managed to get a good result using Anima inpainting.I always have to make the base one with Anima but if I want to inpaint or retouch something I have to do it with illustrious tools
What is with people still working on krea 2 when qwen image 2.1 releases soon?
>>109848866https://litter.catbox.moe/soubs36ja67h0rbi.mp4It's pretty uncensored too. From a starting frame image with this prompt:The anime girl's clothes slide off, leaving her standing there completely naked while she lets out a soft little gasp.
>>109852606Yeah, you'll probably overfit, but overfitting isn't inherently bad. Just save a checkpoint every 2 epochs. What matters is that the LoRA doesn't break in the early steps. Check Circlestone's official LoRA. If I remember right, the Greg Rutkowski one lists the exact settings it used. I don't think it used Prodigy though.
>>109853138Hope it's like Krea. Just being Z Image on steroids would be pretty sloppy imo.
>>109852630UNET>>>>DiT
>>109853940Uncensored =/= good tits and body, looks to generic
>>109852606Doesn't 1280 instead of 1024 eat a lot more VRAM for minimal gains? Or is Anima just unstable at that res, or not trained on it?
>>109852606I usually stops at 2400 steps>I've been using the bucketing feature to lower the resolution to 512 x 640 or 640 x 512This might be why my faces are so bland and un expressive...
>>109853960I dont really know what epochs are but I have 2100,2400,2700,3000 steps savedWhich one do you think will end up better. I'm testing the 3000 steps now and idk, it looks sharp, but I don't really know how overfitting looks.Also, is it really good to train on so many images? Yesterday I trained for testing a lora for the same artist. I used 125 imaged instead of 220 and I had put 2400 steps, prodigy and I think 0.00002 or something instead of 1.0 and for some reason even if it doesnt make sense I'm kinda liking the gens from that test lora a little more
Can the anon using minimax to caption share their system prompt?
>>109854250>I have 2100,2400,2700,3000 steps saveddo a xyz plot with the same prompt and same seed, lora strength at 0.8, 1 and 1.2 and compare.
What checkpoint should I get if I want to try krea2?
>>109854250If it's a character lora, run batches with large seeds, with and without the LoRA. If the images with the LoRA all look very samey, with the face locked in and everything else identical, then it's overtrained. You have to balance the likeness of what you trained against creativity.
big american torpedos
LOCAL =SAVEDhttps://civitai.red/models/2805913/krea2-cat-tower?modelVersionId=3338554
>>109855261Neat
Wasn't here in quite some time but i'm wondering: did ani release apache2 anima? :O
>>109855786No he just drunk himself into mental illness
>>109848876>Buzzwole sipping plasmabretty gud.
>>109855786no as always he was bullshitting for attention and because he was jealous tdrussell got a grant from comfy org and not him (as if the comfy ceo would ever forget ani declared he would "get rid of him")
This is technically anime, but we can all agree it really isn't, and it's way more eye candy than actual anime, right? Is there a name for this whole X illustrator style? Calling it anime feels wrong but I don't have a better word.
>>109855795>>109855927Oh, i see :s
and now as soon as he sees this the "ani #1 whiteknight" (totally not ani btw!!!) is going to spam this thread for damage control
>>109855261This model has zero flexibility, it feels like a lora at 1.3 strength.
>>109853138Not really hyped for Qwen 2.1.,Qwen 3 came and went.
>>109855957Tranime
>>109855911Cute :3
>>109856053Qwen 3 was API and many new SoTA API models were also released at the same time, so it got overshadowed, similar to Boogu and local image.
>>109856012that's because it is literally a lora merged into the full model. he could have just uploaded the lora, but no, gotta force users to download 13gb of weights so the dev can pretend he's making a finetune
>>109852606Look at Circlestone's Civitai page. If you're training a LoRA and can afford to wait, messing around with samples is fine, jst don't expect Anima to forgive your mistakes like Krea 2 does
>>109856325Thanks for alerting meDeleting this piece of shit>>109855261Fuck you for wasting my time
Best model atm?
>>109857072Define best
>>109857072NovelAI v5 (complex scenes) or 4.5 (good looking 1girls)
>>109857072Novel > Krea > Anima > Illu
>>109855957NAI v5 is so trained on that look it's honestly funny. Feels like it had to and learn every last one of them to imitate them and put them out of work.
>>109857505krea can't make anime goon troll
>>109857527Why do you keep lying like this? Anons have posted Krea anime porn here repeatedly.
>Ani its alone>Ani is crying
Any model can be an anime model if you stack a bunch of loras on top of it
>>109857573Krea can do anime out of the box, it's just limited to the really famous stuff. And not a bunch of loras, one is sufficient.
>>109857557post one (1) example
>>109857620https://files.catbox.moe/1haf65.png
>>109857599>It's just limited toAh, here we go again. Let me guess, one style lora that probably doesn't exist and one character lora that probably doesn't exist. Add a decensor lora as well as a cherry on top
>>109857662One style lora for a style, and/or one character lora for a character, correct. Which is the case for other models outside the stuff that's well represented on the boorus. Which is a lot.
It's kind of funny how genuinely stupid our trolls are. Engage them in even a bit of argument and they just fold like a chair.
>>109857641>one girlan hero
>>109857620>>109857641fucking owned
>>109857735Move those goal posts, you little bitch
>>109857748if you goon to one girl you are below 18 and shouldn't be here
>>109857702That's what people want though. Stuff represented on the boorus.
>>109857641can a human goon to that?
>>109857827That's what the hoi polloi wants.
>>109857781>>109857877see >>109857748Look man you got caught lying. It's embarrassing but it happens. Just move on and stop coping like this, it's pathetic.
>>109857072NovelAI for anime. local isn't worth your time
>>109857998are you the guy who lied about krea not doing anime porn
>>109858036post krea doing 2girl, you won't
>>109858163I'll take that as a yes
>>109858249benchod
>>109858027Mommy
>>109858163didn't say it couldn't be futa :)https://pomf.space/wJYaXasu.png
>>109858027oil spil chan
>>109858534>no metadataliar
>>109858534nostalgic sd 1.5 nipples
>>109858534i prefer my anima gens desu
>>109858534This is the power of Krea ... aco tier slopwhoa ..........
>>109858551Hold that L bitch>>109858713So do I, to be sure. It's more of a pain with Krea. >>109858721You're coping, I could have just slapped Niji Sweet Spot on it and your normalfag ass would have been all over it.
>>109858732post metadata
>>109858732>copingThe last 2-3 krea images posted were unironically ChatGPT filter slop. You are hard selling the idea of paying $25
>>109858737Does pomf not retain metahttps://files.catbox.moe/3ivm0k.png>>109858745Like I said, you're coping after being caught lying. Just add a lora you like, probably some normalfag Niji shit, and away you go. It makes anime porn, period
>>109848057Yunyun is CUTE CUTE!!!
>>109859070I never said it didn't, dumbass
>>109859084And I didn't say NovelAI is worse, dumbass, the conversation is clearly about whether Krea can make anime porn and the answer is yes it can
>>109859099Nobody suggested that, retard. Getting this assmad that I called your slop aco tier that you're now inventing new arguments
>>109859111>>109857527Imbecile
Janku Krea 2
>>109859070
>>109859223ai slop
>>109859162it's sad that shitmergers are the only ones willing to gamble on Krea while the respected finetuners sit on their hands.
>>109859282KrAnima doesn't need to be finished to be worth it since loras will fill the gaps anyway.
>>109859166Well litterbox is down too so don't know what to tell you. This is an idiotic argument anyhow since your retarded ass can just go on civitai and see tons of Krea anime porn
i love my wife Yunyun she is CUTE CUTE CUTE and very fertile!
>>109854418Also wondering. I still use illustrous, is krea2 just as good with artist tags from danbooru
>>109854418>>109859412Depends on how powerful your machine is. int8_convrot works on my mediocre machine anyhow. And it only knows famous characters
is julien here? i have a question
>>109855957What's the artist or style for top middle?I want it for Anima/illu
>>109859739Pretty sure it's the guy that did the rabbit hole miku music vid
How do naiggers not get tired of default slopstyle?
>>109859841cute
>>109859894You think cattle have a eye for art?
my wife golshi
>>109859395post it with metadata, you won't
i love my beautiful wife yunyun!
>>109861080girls when they're in those days
>>109859223This is NovelAI v5.
>>109861457girls with post partum depression having their first blood after a lot of time
>"people" are now asking for ridiculous amount of buzz for some finestunes on civitaiSo this is how open source dies?
>anon, i'm gonna cure you from faggotry!
>>109861505>just one more kick and anon will be cured from severe faggotry
Anyone know if wise cards work for novelai?my regular card just gets flat out declined, cant wait until they get paypal back but not holding my breath
>>109861917Says it should work if it has 3DS
>>109861957ah good I think wise has a 3DS system; I guess my bank just straight up refuses any 3DS challenge lmao
>>109861558okay, how the fuck did you do that? did you tune on the cartoon specifically, on your own machine? is the background also generate, or did you edit that? what sort of wizardry is this?
>>109861957did she stuff her pet up her ass or something?
>>109862207I'm drunk so I may give shit instructions.I used the ComfyUI termplate for minimax H3 "image and Video" and used an image to start. It's a .41.1 GB workload. I did this shit with 8 GB VRAM and 48GB RAM and it takes about 17 mins unless you want better quality from first frame which I did (.4 is standard and I did .9)
>>109862264so the thing just automatically knows how to do appropriate animation style from the image style alone? holy shit.
>>109862276Must times, yeah. It's really cool. Here's another I did as a joke to test it on my rig.
>>109860045horses are stupid
How does one achieve this level of slop?Is it even possible to do it locally?
