Designed to make manga-like illustrations. Only trained on black and white illustrations.
Plan to update with more style versions in the future
READ VERSION NOTES FOR MORE INFO
Description
Notes
Has a more mature artstyle, something like Chainsaw Man or Death Note, rather than a more SOL "cute" style.
Reccomended to use Anything v4.5 or AbyssOrange2 for the intended results, but can look good with other models as well, just won't look as manga-like
I reccomend using a lower strength, like 0.40 - 0.60 paired with 'monochrome' and 'greyscale' for the best results.
Use mentioned keywords if color starts leaking.
Info
Trained with Trained with Anything v4.5
Image Set: 40
Training Resolution: 512x512
FAQ
Comments (11)
Great idea! Mage will have a look at this some time soon, thank you !
I'm not getting the same results as your pics even when I drag them into PNG Info and sending it to text2img. Using the same model, steps, sampler and seed as one of your example pics, but still the result is off. Yours is a very clean manga with screentone style, whereas mine looks more like a pencil drawing with soft gradients. Am I missing anything that I should be using? Thanks!
match your negatives, its using negative embeddings.
Try another VAE, for example vae-ft-mse-840000-ema-pruned
I had the same problem with other models
What's this style lora, "marvin"?
Looks like it's the OP's alias for his lora here.
THANKS!!
Hello! I want to know on what illustrations did you teach this LoRa? Can I ask you for them sometime? I really like the style, I would like to use it for my manga, which I do with the help of photography and processing with neural networks. And for this I need to train the model along with photos of my models / actors.
詳細を見た所モデルはAnything-v4-5-pruned.ckptとあり、おそらくSD1.5用です。
NICE! love it!
download link is broken