Previous: >>109740154>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
first for anima supremacy
In 3-5 years all of your images will be animated and have sound. We will never have to fap to still pictures again.
>>109754997FGO has some weird things going on that don't really make sense to an outsider
>>109755327cutie
>>109755220I thought ai was supposed to make our lives better
>>109755220Omg cute ~~
anyone remember neta lumina?
>>109755220What are those little cats tho?
>>109755876The Korean company now focus on LLM roleplay. I was there during the bad time, I saw all of the Anima thing happen
>>109755220the color similarity between character eyes and hair, her clothes and her room was an oversight, or a deliberate stylistic choice?
>>109755077Catbox or promtpt?Doing the same but making the sign say "...using NAI" just to watch the local shills seethe.
>>109756094not sure if the prompt would carry over if you're using NAI but its just leaning on sign, arm on sign, hand up, pointing downward to sign
>>109756057It's a real character brohttps://danbooru.donmai.us/posts?tags=nachoneko>>109756005that's her hairpin
Stamina status? Never had a problem. I don't know how some people generate more than 200 iterations per day.
>>109756515Nice gen
>>109756508>cuck statuswhy would you advertise this?
>>109754339>>109754499
>>109756508i hope anons arent using v5 to inpaint. dont see a reason to not switch back to 4.5 for inpainting
>>109756744in some cases I try inpainting in v4.5 but then switch back
People at NAI must be taking retard pills. First the incredible fiasco of how they managed to communicate the limits of V5. Then the incredible retarded way in which they managed to communicate the change to the subscription credits policy. Now they're banning users that have had an account for years.Did the company suddenly got into the hands of indians? Or has the model become so powerful they got paid to destroy the company from withing so nobody can use it? Or are they just fuckin retards? Who knows! For sure it's very sad seeing AI service providers having their tensorart suicide moment.
>>109756744>>109756508you can save a lot by generating at lower steps, like 11. It still gets good results. At least for my use, which is having the base made by V5, then I use vibes on V4.5 with image2image to style the image.
>>109756744I'm generating at 23 steps and inpainting with 4.5 because it adds beter details in my use case.
>>109756862Only 11 steps? That's too low. It's still kind of strange and depressing that with a service like this, one has to do all these kinds of SaaS workflows like the ones you describe just to use infinite gens. 4.5 seems like a very good model to me, and I often get bored of v5 and go back to v4.5 quite often.
>>109756621I'm not so sure. What's actually different from /vcg/ when people there complain about running out of ChatGPT or Fable quota? Seems like the same thing just showing up in a different thread.
>>109757074I do things like that because I'm making a VN and I rely on vibe transfer to keep the style coherent.Also being that I have to keep the style and characters coherent while generate lots of weird, non-porn situations, I really need unlimited images.V5 even if limited has been great. It's fast and it supports natural language, so by keeping the steps low I can get base images that would be impossible in V4.5.The problem is NAI becoming retarded with their communication, it's something I don't like, it feels like a bad omen.
does anyone had any experience turning an LLM into the Anima prompter?ideally, I wanna have a cheap LLM for production without reasoning, which, even for a complex query, will generate the perfect prompt for Anima in 1-2 secondsrn in production i'm using deepseek v4 flash + a prompt that combines popular booru tags + Anima prompt format guideline. but it doesn't always work perfectly (about 8/10 times), and for complex queries with multiple characters this can drop to 1/10the problem is that my current prompt for the LLM requires following a specific template from “select tags”, because without it, it will start hallucinating and producing nonexisting booru tags that Anima won’t understand. and I haven’t yet found a “template-based” way to universally generate complex scenes with multiple characters. the problem is partially solved by using a more expensive LLM or enabling reasoning, but the latency for the end user will increase, and my LLM API bill will skyrocket
shitty example of using nai's latest Pixel Snap feature on non-pixel art, perhaps there's something it would look good on
astolfo is about to have a cheeky lil' wank inside this japanese post officethe question is, are you a hard enough dude to stop him??
>>109755283>>109754997Fate/Fake gave us Jack the Ripper (theoretical), so he's a shapeshifterKyojiri loli Jack the Ripper (actual killer) was a nameless orphan just trying to survive, but because she was nameless, you get an amalgamation of souls of ALL the orphans (and unborn children) from Britain during the Great DepressionSo whoever summons her defaults to "Mommy"
>>109757465>Kyojiri lolijack's just a loli with a pear shaped body, her ass isn't even that fat
>>109757488Most artists agree with both of us
>>109757217Anima isnt that good to natural language to automate to a llm yet
>>109757258Cool , upscale by near exact but nai
>>109757400>>109757488Why not skin fang but real fang?
>>109757089What's she supposed to be studying here?
>>109756833And how do they buy Anlas if not on the website? Strange. There are plenty of Chinese NAI proxies, thousands of Chinese users on NAI, and even forums dedicated to splitting Opus subscriptions between 2, 3, 4 or more people, so none of this surprises me.
>>109757737>>109757715>>109757705>>109757691Engagement mode: on
>>109757740Guilty as charged. Wanted to make sure no comment went without a reply, just as a bit of a meme.
>>109757217No local or SOTA model exists that doesn't hallucinate danbooru tags, sorry anon. None of them actually know the danbooru lexicon, and on top of that danbooru tags keep changing over time. What you can do is ground something like DeepSeek with danbooru tags as context, but that eats up space, makes the model dumber, and you still risk it hallucinating anyway.
>>109757705if you add the skin fang tag, you'd get oneI prefer the look of pure teeth like he's a dick sucking vampire
Smoking club
>>109757876cigars are hard to genshould prompt it as a thick stogie
you don't really care about music.. do ya?
>>109757922that bulge..nice
>>109757960nicotine and nicotine products cause ED
>>109758090erotic food posting? again?expanded 2.9B aesthetic kinda sucks tho
>>109757826I don't need non-hallucinating LLM, i just need some universal guideline for Anima prompts, that LLM without reasoning will understand. as i said, it's already good for writing simple prompts, but my goal is to turn the prompting complex scenes into an algorithm, so that the LLM can simply follow clear instructions
>>109757992animaexpanded2.9B-aesthetic variant>>109758132sounds like what you need is a character card, then?maybe load them up with top 200 most common tags? and if you can manage to scrape it, 150 most common characters with tags associated with them e.g. hatsune mike, blue hair, pleated skirt, astolfo, hair streak, pink hair, black bow, serafuku, etcLLM then could try to fill in the missing tags, or if you trust anima's tag dropout, no need, but I don't trust itI only do local stuff, so what I would do is write a card in low token count format and load it up in sillytavern and a gemma 4 uncensored running in koboldcpp with a jinja temple telling the system exactly how to respond
>>109758104animaexpanded2.9B-aesthetic+mandatory background
>>109758171Anima understands just a character tag (e.g. `hatsune miku \(vocaloid\)`, i don't need every aspect of the character or more tagsI need somehow teach or make an algorithm for LLM to write complex scene. For example, rn I can write "hatsune miku eats pizza indoors" and it will write a simple good prompt for it, but it won't work with "hatsune miku on the right in black dress eating pizza, while kasane teto on the left looks on her and holding a hot dog". Moreover, I wanna have not only "text to prompt", but perspectively, I would like to make some kind of AI novels with AI photos, and there it would be necessary for LLM to be able to not only write prompts, but also correctly understand the scene from the messagesI personally can spend some time for tweaking my own Anima prompt, I can generate like 30images to get a good one, but I need something that will work for users: instantly on 1st attempt
>>109758216copy paste your hand written multi character prompt, and also one generated by the LLM, I think there's a problem with how you syntax anima prompts
>>109758254the thing is, my prompts with multi character are more random. i can't write a guide or algorithm for writing them, so i can't write prompt for LLM's. Or I just don't know how to write them properlyFor example, i can write to LLM "hatsune miku in swimsuit is eating purple ice cream, close up", it will produce:"masterpiece, anime coloring, safe, 1girl, solo, hatsune miku, swimsuit, close-up, eating, purple ice cream. She holds the ice cream cone near her lips, eyes half-closed in delight, slight smile. Bright summer day, beach background, soft light."I wrote prompt for writing simple scenes like this, but idk how to write complex scenes consistently. For example, LLM with my prompt won't recreate this:"masterpiece, anime coloring, safe,2boys, 2girls, chainsaw mandenji \(chainsaw man\), blonde hair, short hair, sharp teeth, red hoodie, black shorts, holding guitar, electric guitar,makima \(chainsaw man\), red hair, braid, ringed eyes, green track jacket, track pants, holding bouquet,power \(chainsaw man\), blonde hair, long hair, red horns, (straw hat:1.5), white dress, holding cat,hayakawa aki, black hair, blue kimono, holding umbrella,side-by-side, standing, looking at viewer, cowboy shot, simple background, grey background,Four people stand side by side against a plain grey wall, facing the viewer. On the far left, the short blond boy with sharp teeth wears a red hooded sweatshirt and holds an electric guitar by its neck. Next to him, the woman with long red hair and ringed eyes wears a green zip-up track jacket with matching track pants and carries a large bouquet of flowers. Third from the left, the blonde girl with red horns and a wide straw hat wears a white summer dress and holds a small cat in her arms. On the far right, the tall black-haired man wears a blue kimono and holds a closed umbrella at his side."
>>109758363what I do isyear, artist, characters, copyright, as many tags as neededxyz \(fate\) is wearing a pink serafuku with a pink pleated skirt, he has a flat and smooth stomach, thigh gluteal folds are visible, he's standing uprightabc \(fate\) is standing right next to xyz \(fate\), she is wearing a blue serafuku with no skirt on, her thighs are thick, she's looking at xyz \(fate\)if you describe characters, and then describe them interacting in a separate line, it does not work wellnot really relevant to your LLM prompt question but you could try to produce less convoluted promptsmasterpiece, [artist here], 2boys, 2girls, denji \(chainsaw man\), makima \(chainsaw man\), power \(chainsaw man\), hayakawa aki, chainsaw man, blonde hair, short hair, sharp teeth, red hoodie, black shorts, holding guitar, electric guitar, red hair, braid, ringed eyes, green track jacket, track pants, holding bouquet, long hair, red horns, (straw hat:1.5), white dress, holding cat, black hair, blue kimono, holding umbrella, side-by-side, standing, looking at viewer, cowboy shot, simple background, grey backgroundFour characters stand side by side against a plain grey wall, facing the viewer. On the far left, denji \(chainsaw man\) with sharp teeth wears a red hooded sweatshirt and is holding an electric guitar by the neck. Right next to him, makima \(chainsaw man\) red hair and yellow eyes wears a green zip-up track jacket with matching green pants and carries a large bouquet of flowers. Third from the left, the blonde power \(chainsaw man\) with a wide straw hat on wears a white summer dress and holds a small cat in her arms. On the far right, the tall hayakawa aki wears a blue kimono and holds a closed umbrella at his side.
>>109758363and ideally, as more queue items come off with this prompt, seems like you should omit guitar holding tag, and the cat tag, and describe them only in natural language
>>109758447your miku is fried, use lower cfg
>>109758447with slight tweaks for the guitar, poorly described the orange cat>>109758463its the no artist tag anima expanded look
>>109756833they're running out of money and desperate
>>109757400Why the post office though?
>>109758630ideal spots for having a cheeky lil' wank
prompt idea taken from NAI and heavily modified for anima
>>109757751(You) economy is a fragile thing, you can help the market, like the (you) FED
>>109756693Even so, these models are dumb. If you try to make OCs or generate non standard situations, even NAIv5 fails, not as badly as other models, but it still fails at personalized styles or rendering clothes or things for OCs. Diffusion models are semantic sampling machines, they aren't intelligent, they just work with what they were trained on. Those ZZZ girls with hairpins look great because the model was overtrained on them specifically. As soon as you push into OCs or unusual poses, it doesn't take much before the defects show up, gaps in the model's knowledge where it simply doesn't know what to do.
>>109758677Which is the modified idea? Is that tay-tay?
>>109758171>>109758104>>109757992Oh my astolfo my so productive femboy testing things :3
Is Taylor Swift anime? She is very cute.
anything better than anima?
>>109759087KIMI TO NATSU NO OWARI
>>109757955I wouls care but i'm to lazy to learn other thing that isnt dan booru
>>109756515nice! A mecha gen, those are always appreciated on /ad...>schizo rentryI get that you come here to play cat and mouse with Ani, but it's kind of sad that this is the only way you know how to interact on 4chan.
>>109759482
>>109759138Txt2img KreaInpaint like a troon Anima/Illustrious
>>109759506Why though?I have a point. Ani and Catjak are like an immature, dysfunctional couple who use this silly drama to keep their relationship on life support. They both know that without it, there's no reason to keep it going.
>>109759482>>109759548You're not being stealthy.
