Support my work on Patreon and Ko-Fi and get access to tutorials and exclusive models.
Rate and leave a like if you enjoyed it, and follow for new models!
This time I wanted to spend some time and create something different. This LoRA was trained on +1500 screencaps of the anime Highschool of the Dead, plus a few additional images (~100 total). Let me know if you like this kind of LoRA.
The LoRA applies the style even if you don't trigger a character. I recommend using high res.
Characters:
1st row: miyamoto rei, busujima saeko, takagi saya
2nd row: komuro takashi, marikawa shizuka, hirano kouta
3rd row: minami rika, koichi shido, takagi yuriko, maresato alice
Description
FAQ
Comments (22)
For those who downloaded the file immediately after I posted: I added the wrong epoch, which doesn't match the images posted. The correct version has now been uploaded.
dude, this is great
even some lesser known characters
thanks! yeah, I thought it could be interesting to use other characters and even have them interact
Always nice to see more people venture into character packs
thanks!
and it's even more than a character pack, you can create shots that resemble the anime, since I also included backgrounds, random characters, zombies, etc. :)
its very good to include zombies. anyone figured how to make zombie version of the charactors?
@wuhantutg895 hey, yes it is possible, but because the model wasn't specifically trained for that it does take some effort. The model knows what the tag zombie/zombies means, so you can add "zombie" to the prompt and increase the tag's weight, e.g. (zombie:1.3). You can also add other tags that might be useful: blood, blood on face, blank eyes, white eyes, grey skin, etc. You can also increase the weight of these tags, e.g. (grey skin:1.3).
It gives great results! Excellent work! Hope you do more like it.
thanks! yes, I'm thinking of doing another full anime in the near future
@psoft I'd love to see someone do "Daphne in the Brilliant Blue", "Photon: The Idiot Adventures" and "Sexy Magical Girl Ai (OAV)". None of these are on Civi yet, hope you like one of them. :)
Any tips on making a dataset of screencapped images for a LoRA? I'm currently downloading unsubbed videos of Triage X.
I manually captured screenshots that would be relevant for the lora. The total was 1500 screencaps across the 12 episodes + OAV. I then ran an autotagger, separated the screens in folders per character, one with backgrounds, and another one for scenes with multiple characters. I then tagged the caption files in each character folder with the trigger word using BooruDatasetTagManager. Hope this helps.
@psoft Thank you! I'll try it out. Ty for your contributions to AI LoRA making. Your work is amazing :)
@Kazoo_ thanks! glad you like it :)
This is such an excellent quality lora!!! It made my wish come true seeing possibility of HOTD fan artworks using AI generation..
Just one thing I was curious about, Is it possible to also have other characters that did not have a relatively long appreance also added to the lora, such as Misuzu Ichijou?
Pretty sure it would be even better to add these side characters as well. It was kinda shame seeing most of these characters not having a too long survival plot in the main anime&manga, and turning into zombies so fast. (some also apperaed as zombies though, would love to see these as well into lora).
https://highschool-of-the-dead.fandom.com/wiki/Category:Other_Characters
Anyway 5/5 if max is 5, would love to give 10/5 XD
thanks for the feedback!
It wouldn't be very easy do to characters that I haven't explicitly captioned for, but it may be possible to prompt other characters (or very similar) if the description is accurate. I only focused on characters that were more relevant in the anime story line. unfortunately, it's not easy to add to the lora: I would have to take more screencaps (if available), tag appropriately, and run the entire training again. so it's quite a big lift :)
@psoft gotcha! really hope the lora adds more characters but your statements pretty much explained everything.. it is indeed hard to reorganise lora and such. My apologies for the request :)
@nwxwjcrqdcbphdgo8412 no problem! it's just that's it's too much effort for little return :)
I spent quite a few hours to create this one already, with all the screencaps, manually tagging hundreds of files, and adjusting training parameters. and in terms of interest/DLs it pales in comparison with some NSFW models I made in under 1h :)
Hey man, you have discord or any social where to chat?
hey! launching my discord a bit later on. for now you can reach me on this username on Civitai's or Unstable Diffusion's discord
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.