1/1 strength wise seems to work fine. For I2V 1.6/0.8 too, so play around with strengths and let me know.
Example prompt:
She sucks on his thumb with his hand on her cheek.
For more see the txt training data attached on low version.
Description
FAQ
Comments (4)
Thanks for this. Works with I2V as well if you add "An arm enters the frame from the side/bottom" at the beginning of the prompt.
This Lora is awsome! Thx! I use it for I2V with 0.8 on both high/low. Works great.
insane quality. I was expecting some affections on character loras since the act is around a face but it didn't much affect the face. did you blur faces in your dataset? if not I assume you had lots of clips?
18 clips, cropped to only include the face/hand part, no blurring or masking. 12-97 frame length. Balancing wise that certainly worked well here. (txt file with training captions for those clips atleast are attached on one of the LORAs)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.