Preview images don't use any other LoRAs
NO COMMERCIAL USE
I'm not dead yet... This still sucks but maybe I shouldn't clutter titles... It seems to do half-closed eyes surprisingly well despite there being only one mask with half-closed eyes.
"Thomas and Friends" model series are not stop-motion!
It's the closest I can think of to describe the model series style. I'm not calling it live-action because these are miniature engines in an artificial world.
I was curious to see how characters with a much smaller range of expression would work + how this art style would work. Also I wanna see what I can get if I try this together with the CGI LoRA at some point... But the main reason is that I just wanted to try Donald and Douglas with a different flavor, lol
I still don't have a good idea how to prompt T&F characters.
DATASET
86 images total, mainly screenshots from multiple episodes and some other photos. I didn't use materials for Series 12 episode Saved You because S12 used hybrid animation.
I did color correction for images with yellow tint (mainly from episode Gordon and the Engineer).
There are some derpy eye expressions in the dataset, it may reflect in the outputs.
ISSUES
General deformities.
Can't do rear views or snowplows.
Pony
Sometimes it doesn't generate anything beyond the smokebox.
Noob
Ugly results with base Noob. Noses are less accurate.