Let me introduce you to a character you've probably seen before, but not seen enough - Val.
Mostly known for and seen in Red Flags and Mixed Messages MV's
Why I made it:
Because I was blessed enough to watch the musicvideo and although a centipede-themed wedding is out of the question, I dig her style!
Weak points:
Most models aren't perfect. That includes mine. Maybe it's a skill issue. Maybe it's because piercings are just difficult to deal with. Good luck.
Special thanks to
Tom Cardy for his amazing music!
Gabriella Antali for her amazing work in animation and design.
FreckledVixon for the anima version
How did you make it?
~54 total training images from the music videos
~Onsite-training with Flux and Pony running multi-model.
~Illustrious-attempts failed because of skill issue. Redid current with only the Red Flags dataset previously collected
Changelog:
Uploaded illustrious model
Removed prompt/settings from description, read individual version-card instead
Skipped doing v2
Anima version added
Description
Recommended Prompt:
image of tc_val, featuring her beautiful green eyes, pierced lip, pierced earlobe and crimson hair in a ponytail with some hair in front of her face
Recommended Settings:
Steps: 25
CFG: 3,5 - 4
Model weight: 1-1.5, where 1.5 is for full style transfer
FAQ
Comments (5)
turned out fantastic!
works perfectly brother
I also love her look and the style of those videos!
And i want to keep getting better on making loras that contain characters and style i could try to make one for those videos since we have 3 "major" characters there.
And maybe some insight i gonna get could hel You.
btw. yes, AI or at least pony/autism mix etc. really struggles with piercing
Buttons would be easier for sure, as he doesn't have a piercing and there's more art available. Not sure if her dad has a name though, but I also haven't dug that deep. I think for v2 of both models, I'll do them individually and with more specialized tech or at least stronger for character learning, such as higher steps and/or epochs
@DazMakeArt https://civitai.com/models/775207 V1 done, after just training images looks promising