Mike DeBalfo is a Phoenix, Arizona-based graphic artist originally from New Jersey. He has established himself as one of the premier top-selling cover artists in the comic book industry. Known for his pinup art style and dynamic illustrations, DeBalfo's work was discovered by Raven and has since gained widespread recognition. He maintains an active presence at major conventions like New York Comic Con. His distinctive artwork and engaging personality have earned him a dedicated following, with his official page attracting thousands of fans.
Description
Comments (4)
I tried one of the examples, and for me, the LoRa didn't work as expected or as suggested by the example. Unfortunately, the examples don't specify which checkpoint was used. While LoRa might produce the shown results for a Comix-style checkpoint, Atomix Flux GGUF - Q8 only yields semi-realistic images (see below).
This is a common problem. Loras trained with the standard Flux model generally have poor adaptability to other fine-tuned models, whether it's a style or a character lora. I've tried more than 20 fine-tuned models based on Flux, and only two were adaptable. So I suggest you switch to generating images with the standard Flux model, as the results will be much better.
@karl1688
Ok. However, it would be helpful if you mentioned the model you used in your examples.
@Samtiger_ger These two:1, https://civitai.com/models/1076142/flux1-dev-fp811gb 2, https://civitai.com/models/1223425/flux-sigma-vision-alpha1. Flux Sigma can only be adapted to the character Lora, and the other can be adapted to the style Lora.