Lora for the Cake Design we see on the club screenshot preview of Adventure Time: Fionna and Cake Season 2
Lora trained on dataset created from 1 image, if you have interest in a commission like this just send me a message!
No trigger word needed
Character Prompt:
anthro, furry, short hair, animal ears, cat ears, cat tail, cat girl, body fur, white hair, sidecut, white fur, yellow fur, multicolored body, blue sclera, colored sclera, snout, eyelashes, black eyes,
Clothes Prompt:
bare shoulders, sleeveless, necklace, high heels, pink crop top, cleavage, sandals, green skirt, pencil skirt, blue footwear, midriff,
Description
Comments (4)
For only one image, it's very well done! The only problem other than just always looking dark, is when I tried generating her eating an ice cream cone, it generates two mouths, one that is doing the eating is under her snout. Regardless, I wonder how you did this lora?
Hi, I edited the original image, upscaled it, then generated the dataset using a workflow to make extra poses for the character, after that I manually edit the result and feed them back into the workflow until I'm satisfied with the result, then I just train it like any lora.
Do you have a guide on how to do that character turnaround you did with a single image? Looks very interesting!
Hey, I don't have a proper guide, but what I do is generate the dataset with a simple wan2.2 workflow, then delete most images that are to similar (ending with around 20 images), and then training with regular lora calculations.
The main thing is that I need to prepare the initial image properly, that included manually editing so it have all it's details as clear as possible. Sometimes I also generate a "back side" first, then preparing it the same way as the main image and using a first and last frame wan node to generate the dataset.