<hr /><h3 id="issues">Issues</h3>All models and all images in here got flagged as depicting minors, because Persona4 version description said it's trained on Person 4 characters who are teens.All NSFW images hidden automatically.<hr /><h3 id="it-is-time">It is time</h3>It is time I said goodbye to one of my "exclusive" loras that I've been using for a while but never actually uploaded.A thing obscurely named 'devchar style'.DIY datasets already contains a set from that lora under name of 'general anime'. Buuut, to tell you the truth, the dataset doesn't quite finetune Anima into that lora.I was doing further experimenting with Fizgig, and tried to make a tri-slop lora, with chargen, chartact and devchar datasets. But devchar took over - it seems to be the closest to what Anima likes to make.Default output of the lora is very devchar.'devchar-style' - very similar to default, but also seems to escalate face shading issues.'genchar-style' - trained strongly too and replicates 'DIY chargen style' from a long while back.'tactchar-style' - doesn't work.General output is fine, but lora shines with random deepbooru anime prompts and characters. Very pop looking output.Enjoy.<hr /><h3 id="planting-kernels...">Planting kernels...</h3>...with Fizgig.Did some test runs with Fizgig Anima training implementation and merged a couple of resulting loras into some slop. Namely, assorted painting and assorted bara loras trained on anima-aesthetic.It's nothing you haven't seen before, truth be told, and then probably a little less. But it's quirky enough and deserves an upload.As to output - anima doesn't really do trad painting textures very well, so it's sort of has digital softbrush look to it. And you still need to use painting juice to get it. That aside, it's a generic bara lora. Seems consistent from a few tests.Another thing worth noting - Fizgig tags data with natural language. Does it make a difference? No clue.Enjoy.<hr /><h3 id="update">Update</h3>Replaced the lora with a more diluted version.It was a little too rough on the edges and a little below of what I consider pretty or interesting.New version's general output feels a little 'DIY dataset' styley, but '@dizdoodz' tag is a lot more stable in return.<h3 id="dizdoodz-inspired-lora">Dizdoodz inspired lora</h3>A prolific homoerotic artist.Anima is very aware of his existence.I blended it, as usual, for a little more consistency, I guess. But it's still quite raw'ish.(tbf, realisticbeasthero compliments this a lot better, but I didn't want to make another variant of that)You can use '@dizdoodz', if you want to - I didn't train on it, but it exists in Anima and enhances the style.A few backhanded comments on the art style itself.Original art is very refined, neat, clean and sports ridiculous dynamic poses and interactions.But when you start generating basic '1man, standing, flexing biceps' - it kinda looks like most average western-anime mix ever with some flat ink shading sprinkled on top, minus puffed up naughty parts.But hey, at least it's not a "50 shades of hair" waifu lora!Also, looking back, should have probably separated line art from colored art - output is very random.Oh well.Enjoy!<hr /><h3 id="a-few-prompts">A few prompts</h3>Over at realisticbeasthero style I uploaded a small set of colorful dynamic prompts, that I use to generate hunk abstractions. Keywords entirely arbitrary and superficial. They work on some models and not at all on other. But, since I'm uploading abstract images a lot, I thought I'd have them all in one place for anyone to try.Have fun.<hr /><h3 id="japanese-caricature-style">Japanese Caricature Style</h3>First of all, I started training on Anima-Aesthetic-v1.1 now. Don't see any drawbacks, and models end up stronger, I feel, from a couple of tries so far.This model was trained on over a hundred images of expressive "super deformed" Japanese caricatures for 10 epochs. It still very much produces general output and if anything is a good anime lora. For "super deformed" output use your general keyword (that shall not be named).Japanese caricatures are somewhat controversial lately, as such - a declaimer:This model is NOT intended to produce NSFW "super deformed" content. Do not produce it and do not upload your images.Also, I did remix it a little as usual, so it's not too rigid.Enjoy.<hr /><h3 id="2-parts-of-slop-+-1-part-of-kim-hyung-tae">2 parts of slop + 1 part of kim hyung tae</h3>Another lora inspired by one of better unique oldtimer styles.Don't even know what medium his art was used in - never seen any of it. But HUGE thighed chicks burned into my cranium since a very long time ago.While original training data was predominantly female, 2 parts of slop were homoerotic. No idea if any female knowledge left.Didn't work with any less slop, unfortunately. Kim Hyung Tae happens to be a grand master of irl AI-hands. There're almost no images with clearly visible 5 fingers.Still, end result lora is flavorful in its own right.Enjoy.<hr /><h3 id="merge-it-harder-make-it-better-post-it-faster-upvote-stronger...">Merge it harder, make it better, post it faster, upvote stronger...</h3>Remember the oldie Tastefully Nude? Those were the days... Trying to make something out of nothing. Ah...Anyways, I uploaded a new male dataset with a terrible lora trained on it and got visited by 3 Ghosts of Merging.Long story short, I bring you a first pure lora merge (with nothing new added) since like forever.Inspired by a recent series of images made by mixing realistic, hero and beastman styles.Anima merging is quite annoying, but, in the end, I landed on something quite similar.Enjoy.<hr /><h3 id="by-popular-demand!-&quot;a-stack-of-bad-photography-training&quot;!">By popular demand! "A stack of bad photography training"!</h3>Turns out render style lora from Anima Preview1 days is getting attention still!So, I remade it for Anima Base.The original thing contained deconstructron-photography, deconstructron-render art, toon style and a mysterious bit called 'maleabc' (all P1).This one contains deconstructron-render art (P3), toon style (P3) and a brand new 'maleabc' (Base)!What's so good about this?- Glam handpicked sportie guys!- Greater facial variance than normal. Little to no resemblance to training data, but faces are different per gen.- Default realism-leaning style and strong(ish) render style.Biggest con - broken gens are an eyesore on realistic output, and Anima sprinkles errors generously.I do, normally, try to make my loras work at weight 1, but don't forget you can still adjust the weight or mix this with something else, more stable perhaps.Anyways, enjoy!<hr /><h3 id="sometimes-a-prince-is-just-not-enough!">Sometimes a prince is just not enough!</h3>If you know where last one is from - you know where this one is from too :winkyface:Very straight forward.- There may or may not be a two letter keyword that enhances the style (even though, I think this should enhance all American comic book styles across the board through similarity and association).- Conflicting ("word salad") prompts are poor, because data had a lot of "weak signals" in form of auto-tagged comic panel pages. Prompts have to be focusing "strong concepts" for best result (like, you will find that certain sex poses are super consistent, while a random solo dynamic pose will be entirely broken).(for those unaware how Anima text encoder works: Anima uses AI text encoder that infers general "idea" of your prompt and only then looks for latent association; unlike, say, SD1.5 encoder that infers tokens directly in a sequence; that's why the more "about one thing" your prompt is - the better the output in Anima)Enjoy.<hr /><h3 id="something-a-little-more...-charming-today.">Something a little more... charming today.</h3>*bliiiink*Dug up some more pretty pictures in folder named [REDACTED].*bliiiink* *bliiiink* *bliiiink*It's a neat [REDACTED] style, that looks a little like [REDACTED].*blink* *bliiiink* *blink*Signatures show up sometimes, so do use negative prompt 'signature'.*bliiiink*Not much else to add, just a solid lora.*blink* *blink* *bliiiink*Even though lora works as is, you can amplify it with trigger word [REDACTED].*blink* *bliiiink* *blink*Yoghurt is fantastic in this hot summer weather btw!*blink*<hr /><h3 id="it's-ugly...-don't-look-at-it!">It's ugly... Don't look at it!</h3>No seriously, this one is pretty bad.Long story short, with that pixel fix tech I got an I idea I could improve on lowres style lora from a while back. I generated some extra data and made it all neat and pixelated. And it sorta worked. And it's even a stronger than that old lora. But it's not great.The biggest con - for some unknown reason, pixelization shows up stronger with larger negative prompt. So, I'm using a different negative prompt for this one. And I hate it, because I totally wouldn't use anyone else's lora that requires special negative prompt. Oh well.Like before, dataset was generated with SD1.5 Dark and Darker Portrait Lora (<a target="_blank" rel="ugc" href="https://civarchive.com/models/146048/dark-and-darker-portraits-2023">Dark and Darker - Portraits - 2023 - Anima | Anima LoRA | Civitai</a>). I made non-portrait male-centric data. It's strongly stylized, but mixes very well with other loras.- It's pixel art, but unlike the DIY pixel dataset lora, this one is shaded and painterly. In fact, it resembles the origin lora this time around.- Negative prompt in question can be found in my previews.- No triggers required.- Yes, I'm aware there's Anima Dark and Darker lora version too. But I like the SD1.5 stylization.That's it.Toodeloo!<hr /><h3 id="some-people-just-don't-know-when-to-slop...">Some people just don't know when to slop...</h3>A little something from our oriental friends (besides anime).Few tips:- Width resolution of at least 1024 px highly recommended.- Default output is flat illustration. Add painterly tags to get digital art output. Training data was either or.- Try 'asian', 'fat' tags (I tried 'chub', 'chubby', 'musclechub' - these don't seem to work).You're going to take it and you're going to like it!<hr /><h3 id="hope-you-didn't-fill-up-on-cheese...">Hope you didn't fill up on cheese...</h3>This one I hovered for a while but never got it anywhere good.Raw finetune output was almost unservable, as it normally is, when you work with 300x300px images. I had to dilute it - thus the size spike (blended in a bit of deconstructron).It ended up as a cartoony caricaturized model, inspired by VGcats.- OMG! I FOUND IT! THE TRIGGER WORD IS 'TOON'! I REPEAT! THE TRIGGER WORD IS 'TOON'. It stabilizes output by a lot.- You can also use 'sketch' to make it messier and weirder.- You probably don't need any other style keywords. But who am I to stop you.- Simplified hands are very common in VGcats - don't mind the 3-4 fingered output.- Very random, due to source material.- Did I mention model was trained on 1000 images?Testing it myself, but if I find a better blend, you'll be the first to know ;)Enjoy.<hr /><h3 id="is-this-a-thing">Is this a thing?</h3>This bit was in a folder named 'bush' - it's called 'bush style' now.It's a bit narrow and not particularly versatile. Evokes oldie pencil art depicting well-endowed male specimens. Has characteristic rough "geometric" anatomy.Trigger the style with pencil art related tags, like 'sketch'.Give it a spin.<hr /><h3 id="here-comes-a-new-challenger!">Here comes a new challenger!</h3>Um... I had a couple of homoerotic comics lying around, by certain artist named Beastman. Even though google search did not yield anything and only pointed me at furry community - the comics were not furry.Training data was a very good spoof of some famous sfw mainstream anime/manga artists. Resulting lora is a sharp, neat and a "very familiar looking" anime model.Looks like some blend of bleach, jjk and maybe even naruto. Normally, Anima can replicate those quite well on its own, but data was also a little "extra" (if you know what I mean).Enjoy.(still training on anima-base, because aesthetic throws an error; I didn't bother trying to figure it out, but previews are all generated on anima-aesthetic-1.1)<hr /><h3 id="for-90s-corn-lovers-out-there">For 90s corn lovers out there</h3>Lora trained on iconic Julius homoerotic oldschool art.Features rectangular-faced American butch style guys.General style is flat 2d. Painterly tags make somewhat cooler shaded look. Regardless, it's an acquired taste.Enjoy.<hr /><h3 id="another-one!">Another one!</h3>This time it's P4 character art style.If you played it, you probably know, that it's heavily teen leaning, so I had to dilute it a little.Do still watch out for keyword 'young', as output still very skinny twink leaning.Training data was concept arty - if you have trouble with multiple images appearing, do use tag 'solo'.I prefer painterly stylization. But lora does do flat style if you don't prompt for painting.Do use trigger tag. Weighting it higher to try and get a closer style.Enjoy.<hr /><h3 id="amaze-amaze-amaze">Amaze, amaze, amaze</h3>This is a "one step further" version of the recently posted 'chartact' lora, inspired by Final Fantasy Tactics male character stylization.It does contain chartact lora itself plus another mystery bit.Feel free not to use the trigger tag, for more generalized AI look (which is also more muscular). Or do use it to enhance the etching style of Akihiko Yoshida.Good luck!<hr /><h3 id="a-little-watercolor-art-looking-style">A little watercolor art looking style</h3>Anima does already respond to 'watercolor' quite strongly, but making previews I didn't quite witness dat classical wet on wet watercolor look that SD1.5 easily produces.So, I just trained up a little lora on distinct watercolor slop.Turned out quite pretty acshully.(P.S. it even fakes canvas painting texture a little better, if you prompt for it)<hr /><h3 id="crazy-story">Crazy story</h3>So, I went for a walk the other day and noticed a USB stick on the ground.Plugged it in back home, and what do you know! There was a lora in there! What are the odds? 2026 is just full of surprises!And the art style looks so familiar, just can't put my finger on it...Anyways, here it is. Maybe you guys can figure it out.(kinda looks like this I guess <a target="_blank" rel="ugc" href="https://civarchive.com/models/412368?modelVersionId=459604">Artziee - Artziee | Stable Diffusion 1.x LoRA | Civitai</a> , but honestly, seen it all over)<hr /><h3 id="a-little-pixel-art-looking-style">A little pixel art looking style</h3>Okay, okay it's not really pixel art.It's more of a 'lowres art'.Trained on some output I had from this Lora:<a target="_blank" rel="ugc" href="https://civarchive.com/models/146048/dark-and-darker-portraits-2023?modelVersionId=162546">Dark and Darker - Portraits - 2023 - SD15 | Stable Diffusion 1.x LoRA | Civitai</a>Anima can do a lot of art styles, but medium textures are not trivial to land. Strong pixelization is quite fragile. As such, a couple bullet points first:<ul><li>Don't use any medium tags in prompt, (aka 'digital art', 'photography' and such) otherwise pixelization gets smoothed out.</li><li>Faster trained version retains the color scheme of original data. But I opted for a slowly grilled one, that dips into Anima's own pixel ability but uses bright colors. If you want to tone down the color do 'muted color'.</li><li>Seems to be able to mix with one lora, without losing lowres look, but it can vary, I guess.</li></ul>Output is very different from the Dark and Darker lora and looks a lot more general. For better or for worse.Enjoy!<hr /><h3 id="slop-saga-continues...">Slop saga continues...</h3>Anima is so good at absorbing generative AI styles, that I find myself digging through my old output hoard to train some loras.This will possibly be a little series.First entry is a little ink looking style.

Sloppy Moments

devchar-style.safetensors

devchar-style-ESTHET

painterlybara-ESTHET

dizdoodz-ESTHET

caricature-style-ESTHET

kim hyung tae-BASE

realisticbeasthero-style

hyper-style-BASE

realistic-style-BASE

hero-style-BASE

prince-style-BASE

lowres-style-BASE

jiraiya-BASE

vgcats-BASE

bush-BASE

beastman-BASE

julius-BASE

persona4-P3

akihikoyoshida-P3

watercolor-art-P3

hyper-style-P3

pixel-style-P3

ink-style-P3

Files with SHA256: 12f5c761cdbdda513320e76cf1760fac53c4052e2de1b4dfafe3ae11832e281d