This is an experiment to train a LoRA for video, turns out for transformation only High Noise needs training for WAN. Wanted to have something that doesn't have the unwanted motion, pose changes and disappearing clothes of existing LoRAs. Sadly clothes still have a tendency to disappear
Description
A new, but probably final version. I'm through with LTX training, this is hell.
It works ok for image-to-video, for txt-to-video not so much. I recommend a max strength of 0.8