Hello everyone, I’m happily testing video generation in WAN 2.2 model 5B - one of the constant problems is that the video is kind of unstable and twitchy, why is this happening? How can I control the camera more accurately? In what order is it better to create an image (style, object, foreshortening, environment…)
The Wan 2.2 5b model is not good, that’s why. Try the 14b one. For lower ram i would suggest to use gguf in combination with speed loras.
There is no camera control really. There are some loras around, but i am not sure if they work with Wan 2.2. And you cannot define a path or something. You need to try out , give some keywords, toy with loras, and hope the best.
For prompting, there are some guides out there. Wan2.2 Prompt Guide - Cinematic Video Generation Prompting & Control
Hope that helps
Arunderan