Wan 2.2 LoRA of Riley from the first Inside Out, trained from screenshots only.
I trained about ~3.5k steps using images that were auto-captioned by Joycaption, and then another 2.5k steps with just the tag "rileyandersen". It'll work using either "rileyandersen" or "riley andersen", but the former is the primary trigger word.
Even though it was trained on images only, Wan 2.2 out of the box does seem to do really well with animating her, so I almost feel like I don't need to add any videos to the dataset, lol. But I might try that at a later point, if I do another version of this.
