Hellsing Ultimate OVA style -- LoCon for Pony, LoRA for AnimagineXL 3.1

A style LoRA (a huge LoCon for Pony, a large LoRA for AnimagineXL 3.1) for the Hellsing Ultimate OVA series.

A quick plug: I have other classic anime style LoRAs/LoCons. Check out https://civitai.com/user/SweetRammaJamma!

__

SOME INFORMATION SPECIFIC TO THIS LORA/LOCON:

For the Pony LoCon -- I cooked it for 19 epochs. Epoch 17 was the first one that seemed to establish an understanding that Seras's unique hair (which I tagged "flared bob cut") is not necessarily associated with Seras, so I'm uploading epoch 17, because it allows you to do things like "black hair, twin braids, MCSeras," AND you can also apply "flared bob cut" to other characters (although you may need to up the weight:1.2). HOWEVER, this LoCon is definitely overfit in some other ways, just because some features are common on ALMOST EVERY HELLSING CHARACTER, so character trigger words tend to result in features like white gloves, ascots, and high-collar shirts/coats. It doesn't seem to be a significant problem unless you use the "MC" character tags, so, just be aware.

The newer AnimagineXL v3.1 LoRA works fine with the base AnimagineXL v3.1 model, but AnimaPencil v3.1 required far less tinkering for my sample images, and making the images was just plain more fun. Also, compared to the Pony LoCon version, specific characters/concepts might seem overfit or underfit (which is largely due to MY different training settings/approaches, but you should still be aware). In general, if you don't care about the unique features of AnimagineXL, then I recommend trying the Pony LoCon first.

Primary trigger word is HZLZVZ.

USEFUL NEGATIVE PROMPTS: "blurry, blurry background, field of view" does tame some of the "authentic" blurriness of the source material. Likewise, "greyscale, monochrome" can help with some of the source material's washed out colors. "White gloves," "ascot," "necktie," "neck ribbon," "collared shirt" are the sorts of things that can help remove some of Hellsing's "nearly-universal features" WHEN they show up. And, "simple background, gradient background" seems to always help with SDXL anime stuff.

FOR PONY CHECKPOINTS, TRY: 1) ColorburstXL, 2) LuminaverseXL, 3) Reponygine, and 4) "Prototype" (see the "GENERAL INFORMATION" below for more info, as well as "RECOMMENDED RESOURCES" at the bottom). Keep in mind that there's non-Pony versions of some of those checkpoints, so be sure to get the Pony ones. Also, all of my testing is with "score_9, score_8_up, score_7_up, score_6_up, source_anime" at the beginning of the positive prompt, so try that first.

Characters were tagged with the following trigger words. ... followed by additional common tags for that character. General disclaimer: Most of the character tags work quite well, BUT, "specific character likenesses" were NOT my actual goal. Main characters may be overfit because of the sheer number of images. Less common characters are likely to be either overfit or underfit (either because they have so few images, or their images all share very repetitive features like identical clothing, identical backgrounds, etc).

MCSeras ... flared bob cut, blonde hair, red eyes, blue eyes, military uniform, hair over one eye, thighhighs
MCAlucard ... wide-brimmed hat, red headwear, opaque glasses, orange lenses, black hair, long hair, red eyes, necktie, ascot, sharp teeth, fangs, vampire, handgun, formal, coat
MCMajor ("The Major") ... glasses, yellow eyes, necktie, opaque glasses, brown hair, round eyewear, formal, uniform
MCIntegra ... glasses, blue eyes, blonde hair, long hair, cigar, cross, necktie, dark skin, formal, suit, round eyewear, eyepatch
MCAnderson ... glasses, facial hair, blonde hair, opaque glasses, green eyes, cross, sword, stubble, cross necklace, evil grin, gloves, beard, round eyewear, spiked hair, mature male
MCPip ... cowboy hat, brown hair, eyepatch, braid, green eyes, scarf, uniform, cigarette, jacket, single braid
MCMaxwell ... purple eyes, ponytail, grey hair, white hair, cross necklace, priest
MCZorin ... masculine female, muscular female, tattoo, green eyes, orange hair, face tattoo, spiked hair, short hair, body writing, arm tattoo, english text, sleeveless bodysuit, crazy eyes, scythe
young MCWalter ... flared ponytail, opaque monocle, black hair, monocle, necktie, fingerless gloves, formal, suit, round eyewear, light blue eyes, black gloves, black pants
old MCWalter ... flared ponytail, opaque monocle, monocle, black hair, light blue eyes, round eyewear, black necktie, fingerless gloves, white shirt, black vest, formal, pants
MCWinkle ("Rip Van Winkle") ... glasses, blue eyes, long hair, black hair, freckles, rifle, blue hair, round eyewear, hair between eyes
MCCapt ("The Captain") ... red eyes, white hair, military cap, trench coat, gloves, military uniform, muscular, dark skin, dark-skinned male, topless male
MCTubalcain (AKA "Dandyman") ... fedora, brown headwear, facial hair, mustache, necktie, green eyes, black hair, formal, cigar, suit, holding card
MCDoc ("The Professor") ... opaque glasses, blonde hair, labcoat -- maybe also try "goggles" and "midriff"
MCSchrodinger ... cat ears, blonde hair, necktie, pink eyes, short hair
MCHeinkel ... blonde hair, sunglasses, green eyes, cross necklace, priest, round eyewear
MCShelby (Shelby Penwood) ... blue eyes, facial hair, mustache, necktie, formal, old, wrinkled skin, suit, old man
MCYumie ... black hair, katana, cross necklace, nun, habit, priest
MCKathliks (this is how I tagged the Catholic soldiers with the white pointy hoods. I haven't tested this tag. They are always seen in groups, so be prepared for that!)
MCDad (this is how I tagged Pip's dad in a flashback... not because I was trying to capture Pip's dad, but rather because I wanted to try to separate his beret from his character. I haven't tested this yet, though!) ... black beret, green eyes, sunglasses, white hair
child MCWalter (yeah, he's a teenager in the flashbacks, but I just had to pick a tag, and "child" seemed to work to differentiate him from his other forms) ... black hair, short hair, light blue eyes, necktie, vest, fingerless gloves, hair over one eye
child MCAlucard (yeah, he's hundreds of years old in the flashbacks, but I just had to pick a tag!) ... red eyes, black hair, blunt bangs, hime cut, necktie, fangs, hat, fur hat
MCVlad (this is how I tagged Alucard when he was in his... uh... "classic" form. With the beard and armor and sword and such. If I do a version 2, I'll probably tag him with "old MCAlucard") ... red eyes, black hair, long hair, facial hair, cape, sword, hair between eyes, armor, messy hair, gauntlets
MCLukeVal (Luke Valentine) ... green eyes, blonde hair, glasses
MCJanVal (Jan Valentine) ... yellow eyes, dark skin, dark-skinned male, hat, nose piercing, earrings, lip piercing

Finally, you can use "wide-brimmed hat, red headwear" for Alucard's red hat, or even "wide-brimmed hat, blue headwear" to get a blue version of Alucard's hat, and so on. Unfortunately, "wide-brimmed hat" does seem to be "contaminated" with other features of Alucard's clothing (despite every image of his wide-brimmed hat being ALSO tagged with "MCAlucard"), so be prepared to use add stuff like "collared shirt" and "ascot" to the negative prompt. Much like with "MCSeras" and "flared bob cut," epoch 17 was the first epoch to reliably understand that 1) "wide-brimmed hat, red headwear" can be distinct from "MCAlucard," and 2) "wide-brimmed hat" can be distinct from "red headwear." But, epoch 17 does seem to associate "wide-brimmed hat" with neckties, ascots, neck ribbons, and collars -- despite the fact that "MCAlucard" is indeed tagged with "ascot" and/or "necktie" where appropriate. It's probably something that I can debug for a version 2, but I want to move on to different style LoRAs for now.

_______

SOME GENERAL INFORMATION ABOUT ALL OF MY LORAS/LOCONS:

Sorry about the lack of image generation data. I'm going to work on uploading images with generation data when I have time/energy. In the meantime, I can tell you:

To generate the sample images, I don't use outside LoRAs. As in, my sample images only use 1) the main checkpoints, plus 2) my own LoRA/LoCon. I don't use embeddings, either. So the sample images are very "pure/simple."

I use ComfyUI only, so my image generation methods might be different from what's typical with Automatic1111 or CivitAI's online generator. Notably, I don't inpaint! If things get too weird, I just "roll the dice" again! But chances are good that inpainting will be an asset to you, especially for things like getting eyes that are 100% stylistically correct.

99% of my testing is with Euler A, at 25-30 steps, CFG 6-7, with the LoRA/LoCon at strength 1.0, and with the trigger word at the very start of the positive prompt. I test both latent upscaling (1.25x to 1.5x, 0.60-0.65 denoise, ~12 steps) and "regular" upscaling. I've tested face detailers, too. All of it works appropriately, but some typical "light troubleshooting" will be required if you're going for a specific output -- this is StableDiffusion, after all.

I do upscale my sample images quite a bit. Usually, it's latent upscaling first, then an additional pass with "regular" upscaling at about 1.5x. I usually resize with boring old Lanczos, and not "model upscalers" (such as AnimeSharp4X), because Lanczos resizing ends up more faithful to the original anime's style -- but really, I do whatever works.

In regards to upscaling and YOU: Don't be surprised if something like my Guyver OVA LoCon gives you gritty/grainy initial results. Experimentation is required. And it varies from LoRA to LoRA -- for example, FLCL is a nice-looking and somewhat modern anime, so it doesn't have the same problems as the Guyver OVAs.

Where you put the trigger word does matter, and (the trigger word's weight:1.2) does affect things. So to decrease the "faithfulness/literalness" of the style, put the trigger word all the way at the end, and (decrease its weight:0.5). Of course, you can play with the weight of <the actual LoRA/LoCon:0.8>, but I tried to ensure that things work reasonably at 100% strength. (If my LoRAs/LoCons are totally unusable at 1.0 strength, I simply don't release them! I find another way to re-train them!)

For both Pony and Animagine, I try "simple background, gradient background" in the negative prompt. This does seem to help with the typical boring backgrounds of SDXL anime models, and it might help more than usual in cases where a LoRA/LoCon is based on an anime that itself has many boring/simple backgrounds (FLCL comes to mind).


___


GENERAL INFORMATION FOR MY PONY LOCONS:

My Pony testing was all done with pretty standard quality tags in the positive prompt, in the form of "primary trigger word, score_9, score_8_up, score_7_up, score_6_up, source_anime," so I can guarantee that those work, but feel free to experiment!

For checkpoints to combine with this Pony LoCon at 1.0 strength, I recommend starting with the following (which are in the "Recommended Resources" at the bottom of the page):

1) ColorburstXL (Pony), because it produces "faithful style" while still being more consistent and aesthetically pleasing than the base Pony model. Then, maybe try...
2) Luminaverse (Pony) will get you something more "modern" while still being quite faithful to the LoCon.
3) Reponygine (Pony) will get you something quite "modern and pleasant," while still retaining good core style features (like eyes and hair details) and character likenesses.
4) "Prototype" (Pony). I don't quite know how to describe the look yet, but it works well.
5) AnimeConfettiComrade (Pony), IF you want a MORE "literal" interpretation of the LoCon, while still being more consistent than the base Pony model.

You CAN use this Pony LoCon with the base Pony model, but the style might be overpowering, and you may have to lower the LoCon's strength a fair bit below 1.0, and/or move the trigger word to the end of the positive prompt, and/or reduce the strength of (the trigger word:0.1) -- which are all the same ways that you'd adjust the strength with other checkpoints, of course. Adjust to your taste, always :) And, there are many more Pony-based checkpoints that I haven't even tried, so if you find one that works well with this LoCon, please let us all know!

_

GENERAL INFORMATION FOR MY ANIMAGINE XL 3.1 LORAS:

My testing was all done with "masterpiece, best quality, very aesthetic, absurdres" at the end of the positive prompt, and "worst quality, low quality, jpeg artifacts, lowres, blurry" at the end of the negative prompt. If necessary/helpful, I put the primary trigger word at the very, very, very end of the positive prompt, after all quality tags.

The LoRA should work fine with the AnimagineXL 3.1 base model at 1.0 strength or a little below. You can use AnimaPencil if you want to try a 2.5D look, but you'll be sacrificing quite a bit of the overall style.

Let us all know if you find any other checkpoints that work well with the LoRA!

Also, I've heard that AnimagineXL 3.1 supports a shit-load of artist tags, so if you're adventurous, you could try combining with those... but I haven't investigated that yet.

|___

Yes, my filenames are long. It's because I personally use LoRA filenames to quickly compare different versions of my LoRAs while testing things. I recommend renaming the file to your tastes! Though it would make sense to leave the primary trigger word in the filename, right? (It's up to you!)

Yes, my Pony LoCons are extraordinarily large. But I made them for myself, and, to quote my genius friend, "it works for me and what I do." If you try these Pony LoCons, and you've made some SDXL style LoRAs/LoCons that you like, I would be happy to discuss training settings with you.

My AnimagineXL 3.1 LoRAs are also quite large, but more standard. I trained them as regular LoRAs instead of LoCons because I've always had weird results training/testing my style LoRAs/LoCons on AnimagineXL (going all the way back to AnimagineXL 2.0), and keeping things simpler seems to work better. Note that I don't release the AnimagineXL 3.1 LoRA versions unless they work OK, so if one is missing, that's why. I plan on troubleshooting that later, but for now, I'm trying to work through uploading my backlog!

I keep training datasets identical for both the Pony and Animagine versions. If there are any minor differences, I will make a note of that in the "specific information" section closer to the top of this page. (Major dataset changes will be noted by the version numbers, like "Pony-V1" versus "Pony-V2," and "ANXL3.1-V1" and "ANXL3.1-V2")

__

No warranty, express or implied.

I place no special restrictions on what you may do with these .safetensors files. Likewise, I am not responsible for anything that you do with them.