A separate KisImage for every frame is a nice idea, but we need to answer two questions:

1) How would we handle static layers? Are they going to be duplicated in every frame? Right now all the types of layers store a pointer to the image (KisDefaultBounds in every KisNode + a direct pointer in every KisLayer). Also clone layers have a bidirectional link with its source, which is used for projection calculation every time. Every paint device inside a layer also stores a pointer to KisImage via KisDefaultBounds.

These pointers were the reason why I decided to implement LOD by switching datamanagers inside a paint device, instead of switching paint devices in layers.

2) Memory management. All the duplicated node will occupy extra memory. Krita's tile engine has copy-on-write mechanism, but it works only on direct copies of paint devices. If we save the duplicated nodes into .kra and load them afterwards, the memory will be duplicated.

---