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!
LoRA inspired by the style of the artist ama mitsuki. Follow and support the artist if you like their work.
No trigger word necessary. As usual, I made the lora to be used at 1.0 weight. But, as with any style lora, the effect may depend a lot on the base model you use and the weight of the lora. I recommend using hires fix / upscaling.
Description
FAQ
Comments (6)
How to reliably achieve the multi view style?
base models typically recognize the tag "multiple views". you should also be descriptive of the character/outfit.
I am able to achieve some cool things with this Lora alongside with others, but really hard to control.
@pigggggggy the "multiple views" should work when prompted, but probably it's hard to get the right representation that you want, right? that would probably require thoughtful tagging of the dataset, but even then SD is unlikely to cooperate on the exact placement and composition, unfortunately.
thanks for the feedback!
Many thanks for your great work! Amamitsuki is one of my favourite artists and the picture samples that this LoRA produces are very adorable. May I ask what training parameters you set? There may be some small aspects that would improve the likeness of style, in my humble opinion, you might as well make these girls look slimmer, taller, and add some emphasized, expressive high/low-Angle views, but it is just a matter of preference and of course can be achieved by some other ways. Looking forward for your updates.
@Volturipper thanks!
~1000 images non-cropped, 512 res, 0.0001 LR with Adam8bit, 10 repeats, batch size 16
but I don't think the params matter that much. I think removing certain images from the dataset could improve a bit
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.