The LoRA is designed to create a sudden “blink-cut” scene change: a character who is initially facing the viewer head-on abruptly shifts to a different pose and framing, where the same character looks toward the viewer.
The exact prompt phrasing for the transition is provided in two variants, depending on the target posture:
Example Prompt (facing viewer)
Example Prompt (smashcut)
A more detailed prompt description is within the ZIP file.
Description
consistent scene change into "whatever you want"
These LoRAs work best on images with wider aspect ratios and produce body horror about 20% of the time, despite my best efforts. It took me about 5 full training runs on the high model to figure out how to get somewhat consistent results.
workflow: Wan 2.x img-to-vid: https://files.catbox.moe/0ssyip.json