<My first lora on here.>
Works well with I2V or T2V.
I trained on the movie clips (36 clips trained at 768 res from cropped 4K footage) as well as still frames and some AI images for the...ones that weren't in the films.
Say: "wearing a blue police uniform"
"with floppy ears" will help fix that if you get 4 ears or if you prompt her wearing a hat.
"wearing a blue hat with orange vest"
*rank 64. 7000 steps
Wan2GP was used for all examples generated with Distilled LTX2.3.
Description
This is 5000 high noise steps and 2000 more steps on balanced.
I messed up some captions and you might see a random Nick Wilde every now and then, just ignore him. I'm going to use EditAnything to get him out of all my datasets and he will be a ghost soon.
This is trained purely on movie data, no AI images because those nudes hurt the likeness too much. Now she looks like good ol' Judy as she should.
Just use AI start images if you want her naked.
FAQ
Comments (2)
Yea I have a physics enhancement lora that I trained to give videos a much better sense of physics interactions but civit is so buggy I can't really reliably release it. Have not seen download numbers update for any creators models for days now, pages are bugged out and when I saw the other version of your lora show up in the feed I clicked it and it was 404. Last thing I want to do is release it then try to figure out why no one else is able to see the model but me for hours (happened right before the current buggy shit) or deal with the model just deleting itself, which has happened before and wasting my time. Pretty damn annoying honestly. Anyways, I trained the new lora on the Ltx2.3 Dev model merged with my reasoning lora it feels like it made the lora learn more than it would have than just training on the normal base model imo. If you want to experiment with training on it lmk and I can upload it to huggingface.
Please consider loona wolf (the sexy version) next if you are planning on training another lora.