>>22518

A lot of the weirdness has to do with how Grok does extensions. It automatically carries forward a lot of information about the first video (scene, tone, people, motion, camera movements, ect..) so extensions tend to be resistant to adding new elements to the scene or sudden changes to tone or style because it basically sees it as conflicting information. It can help to start an extension prompt with re-enforcing the continuity of the previous scene and THEN describing how the scene changes but my preferred method is using extensions only if the scene is continuing as-is. If there's going to be a sudden shift I'll generate the first clip, then take the initial image and the last frame and ask Grok to write a detailed image edit prompt to turn edit the initial image to the last frame. Then I'll use the newly transformed image to generate an entirely new video and just stich the two videos together and add a fade or some other transition between them