Disclaimer: Some preview images were obtained at different stages of the training process and will not match your output 1:1. They are here to give you an idea of what this LoRA looks like.
A dataset of 1105 assembled sprites, each downloaded painstakingly one-by-one and denoised through a batch converter, were used in the making of this LoRA.
Compatible with other character LoRAs.
Recommended tags to use to replicate Arknights' sprites:
chibi
full body
cel-shaded
flat colors
arknights-inspired clothing
Recommended dimension ratio: 1:1
512x768 produces taller chibis.
Lower/increase weight as needed. Works at fine at 1.
Enjoy.
Description
I found that the first epoch had an adorable style to it so I might as well throw it here.
FAQ
Comments (4)
What model was used to train this LoRA?
I think it was Anything 4.5. But I'm not the creator so I can't say for sure.
i'm kind of new to generating ai art.. everything i try and make with this comes out unrecognizable and completely different from what i describe, even when i follow the exact same settings and prompts from the sample artwork. Any tips?
I'm also pretty new, but I've found the best way to check a new setup it to download the example image and use the PNG info to get the settings and copy those to txt2img. Then, double-check you're using the correct checkpoint, and check the prompts for any loras you may be missing. Importing data from PNG info does not set your checkpoint to the one used for that image, so you have to make sure its correct. Then, if a VAE is used, check to make sure you're using that VAE, which can be set in settings. If the image has a provided VAE, made sure youre using the one that image used, sometimes the one you download is incorrect. Hope this helps!