This v2 should finally work better with full body shots, but keep in mind this model is mainly thought for medium shots, portraits and closeups. JoJo faces have way too many expression lines, so SD messes up trying to simplify them. With high resolutions, it should work fine, though.
Trigger: 'Jolyne_Anime'
Optional tags: 'coat', 'pants', 'face focus', '(((pink hair)))' -> (for the "alternate pallet" effect)
Weight: [0.6 - 0.9]. Go high for a strong art style (with less flexibility), and low for creative uses or poses (and even lower for realism). With closeups, the art style should remain consistent until around 0.65.
Suggestion: Experiment with lights, colors and visual effects. Some results can be quite interesting!
Checkpoints used in examples: 'Anything V5' and 'UHD-23'
______
Example prompt:
Jolyne_anime, confident, ready to fight, creative shot, green eyes, green lips, perfect quality, (absurdres), 4k, masterpiece, lens flare, light rays, (psychedelic background), illustration, ((anime screenshot)), professional lighting, posing for viewer, ((psychedelia, face focus, wind, detailed face)) + [LoRA set to 0.65]
Description
Better full body.
FAQ
Comments (15)
what model did you use ?
Hey! I used Anything V5 and UHD-23
But I think it should work fine with most anime-focused checkpoints and flat coloring styles
Sorry, but where are your Pokemon LoRAs? (May ORAS- May anime, etc) I can't find them anywhere.
Hey!
So, long story short: A group of users started review bombing all my Pokemon models (I don't even know why), and easily tanked their scores and even my user rating. I got tired of reporting and ended up taking everything down (this is a new account).
Making models takes time, so not gonna lie, those things take the fun away from using this place!
If you lost any model, I can always try to get it to you somehow, but I don't think I'll be reuploading them or making new ones soon (Jolyne was already finished, so it felt like a waste)
Sorry about that!
@koop_ Thank you for replying! Sorry to hear that, there are some weird people out there. I really loved your PKMN models (are you gonna still update them, even outside Civit?). So what I got from you are:
Dawn - Pokemon DPP [Anime Design] v1
May / Haruka - Pokemon [ORAS Design] v1
May / Haruka - Pokemon [ORAS Design] v2
May / Haruka - Pokemon [ORAS Design] v3
May - Pokémon RSE [Anime Design] v1
Serena - Pokemon XY [Anime Design] v1
Do I miss something?
Thank you again for your reply!
@Gaspatcher Hey, no need to thank! Actually, happy to read you liked them! :)
Pokemon-related… those are about all! I was working on a Serena v2, but still not sure if it's overall better than v1 (more consistent art style and clothing, but less flexible).
It's a shame how easy it is to mess with creators with the current review system. Almost no one writes reviews, so it only takes a user and a couple of alts to target someone.
You know what's funny? It all happened after I offered to make a model of Ash because a user requested him, and I think some people may have got nuts because of fandom shipping crap I don't even care about (I don't even watch the show lol). Of course, it may just be a coincidence but… (facepalm emoji)
Anyway, thanks again for your interest!
@koop_ I don't mind if it's less flexible, if you upload v2 of Serena, I'll take it :) Just tell me where you can drop your LoRAs (it doesn't have to be on Civit)!
Yeah I think I saw some nasty comment when you considered doing Ash with other girls and people got angry (they're way into shiping) - I'm afraid it's no coincidence...
@Gaspatcher Sure, I'll let you know! I've found other sites outright stealing my models, if I can claim them, I'll update the Serenas with the v2
But if Civit were to make something about review bombing, I'd prefer them to be here!
(PS - Seriously considering creating Ash anyway just to bother lol)
Hi ! it's a cool lora ! i'm learning how to train lora, wich settings are you recommends for good results as you ? : )
Hey! Nothing out of the ordinary, I mostly use the basic/recommended settings for anime LoRA, but the real work is with the image selection.
I always tend to use a lot of images (over 300) trying to cover both body and face from the most possible points of view. That way, SD doesn't need to "guess" how a model should look when posing, as it will already have a reference (that's why some LoRA sometimes look bizarre or generic when changing the default pose).
Apart from that, I always use pictures with the same artstyle. So, no fan art! With Jolyne I basically used frames from the whole season looking for poses and parts of the body I didn't have covered. And then, I chose a specific episode where the animation looked gorgeous to take detailed screenshots from her face, eyes, tattoos, clothing details…
Hope it helps!
@koop_ Thank you very much for responding. I think that was my mistake, because I selected pictures in different positions but with an art style that did vary, I will be more careful next time. Thank you very much ! By the way, how many steps do you use max and epoch?
@TheTataAI No worries! About the steps, it really varies... if I have a lot of good and unique images, I usually go for no repetitions and end up with 3000/4000 steps (sometimes even 6000), but keep in mind that if your images are very similar (a lot of similar shots with similar lighting, same face expressions...), the LoRA will most probably be overcooked. In those cases, just aim for the standard 2000 steps by removing redundant images and playing with the repetitions.
About epochs: With no repetitions and a good image selection, the last one is usually fine for me, but with repetitions (and thus a limited image selection), I'd recommend to make 10, keep the last 3 ones, and check if there's a real difference between them. Basically, the lower epoch that works, the least overcooked the LoRA will be.
@koop_ I am very grateful for your advice, you are the first person to answer these questions of Lora, Thank you very much. I will be doing trial and error to see what I can get. I will follow you on this platform and see all your work.
@TheTataAI Happy to help! Just keep practicing, work a bit on the image selection, and you'll see great results soon!
This model is great, not only does it have several outfits, it also has the different color palettes of the anime, I would like it to be updated to Illustrious
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.