LoRa made to resemble the art style of HADES and HADES 2.
The dataset was entirely made with character art from both games with white backgrounds, so it's best for that kind of images and probably won't do much for backgrounds. It requires no trigger word and I've had great result with a Strength of 1.
ERRATA: I do realize the file name says FLUX, that was a mistake I made when setting up the training but the Lora IS MADE FOR IL so please don't mind the name and do let me know if it can be retroactively fixed.
Feel free to share your creations of whatever kind below, I'd love to see your work!
Description
FAQ
Comments (13)
genuinely great work, looks very authentic.
Just asking but does this have any hades2 bathing scene artwork, the dark aesthetic with green highlights looks also very nice.
Thanks! Sadly I trained this using only background-less pictures of characters (the ones you see when talking to other characters in the games) so I don't think it's going to do much for backgrounds.
the filename says FLUX. which is it?
It was trained for Illustrious, the flux thing was a typo I couldn't fix in post sadly. But it's nothing more than a name, don't worry.
cool thanks its awesome
Actually it's more than just a name thing. Let me explain.
I've looked at the architecture of this LoRA and layer structure strongly suggests FLUX arch, while other parts point to SDXL/Ill.
Early transformer blocks are okayish, text encoder attention layers are partially messed (some overlap), channel dimensions are mismatched, there are different normalization schemes and attention heads that don't align.
In other terms, it's a very interesting LoRA and while weight block shifting is kind of impossible here, normal usage (static strength over sampling steps) is acceptable. Due to potentially good dataset and real strong concept adherence it is very interesting, but for power users it's complete mess and artifact factory.
Don't be offended though, everyone makes mistakes and learns from them.
This is (again) a very interesting LoRA, it's possibly even better for FLUX than it is for SDXL so people can experiment with it and get great results.
On a side note, it is by far the best "style" from all the other Hades LoRAs on civit. :)
Is persephone in the lora?
I used all the available characters from Hades 1 and 2 for the dataset so it should be pretty easy to accurately replicate her with the right prompt. Sadly, the lora isn't trained to recognize individual characters by name, just the style itself, so I doubt adding "Persephone" to the prompt will do much, haven't tried it myself tho.
tld:dr She is, but only as part of the training data
CeUnoXingqiu i have tried, blonde hair, hair over shoulder lavender, hair flower, but no luck :(
Works perfectly well, awesome tool.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.