Why don’t you do the whole parenting/dissolve infinite zoom?
I think videocopilot.net has something along those lines. Then you can scale animate 1k or 2k images, and dissolve them, which might complicate your project, but is much easier to work with. It makes much more sense and is much more efficient than zooming a 21k image. It’ll also save you plenty of render time.
Nicholas Toth
Freelance Animator
nicholastoth.com