I’ve been experimenting with Deforum in Stable Diffusion. Using the Ink Punk Diffusion model, I rendered a background and the foreground separately and composited them. I could better replicate the camera movements in this shot using Deforum, but who’s got the time right??
I’ve been trying to figure out how to get good video effects using Deforum, but it’s been a real challenge. There’s a tension between the quality of generation for a single frame vs stability of the image over time. The composite approach has been the best strategy I’ve found to balance them, but that’s still limiting.