After four iterations of this LoRa, this will give you something resembling the character Felix the Cat using only the given "Trigger Words" (#TW) as a prompt.
This how the cover image for the model was made. I've also been able to add additional information to the word prompt and get pictures of "Felix Riding a Bicycle" and doing other things. However, I find I always get better results when using a starting image. Also, I find that character focused LoRa tend to be a little janky. I think that's just because it just takes one ~little feature~ being a ~little off~ for the result to feel totally wrong (when it comes to recognizable characters in particular).
I think all versions seem to work better when used on their own. In my experience, the results tend to look more wonky when you use two or more of the LoRA (for the same character) at the same time.
