Currently ‘replace duplicate frames’ only works with frame interpolation (increasing frame rate or slow motion) but what if we only want a video to fix duplicate frames and keep original framerate ?
The idea is to make the duplicate frame workflow in TVEAI a standalone 2 passes feature like Stabilization.
The program would first scan the video to find and mark the duplicates frames then a json file or whatever is saved for the second pass where the program will then delete and replace the duplicate frames with an interpolation model (Chronos seems more suitable for this task than Apollo that internally interpolates 8 frames)
And only the replaced frames would be processed, the other frames would just be directly copied keeping them untouched. This would allow the second pass to be very fast.
And an advanced feature would be to scan ahead and analyze the motion of a few frames after the duplicate(s) frame to determine where the interpolation should really happen because on many videos with duplicates, like variable framerate ones converted to constant framerate, duplicates frames an dropped frames are generally random and placed to keep the soundtrack in sync.
So at the end of the 1st pass a file is created to tell the AI what frames are to be removed and interpolated
something like this
find duplicate at frame 20, find likely jump at frame 23 >>> remove frame 20 and create a new frame interpolated between frame 22 and 23
There we can add a manual slider to tell how many frames to scan ahead to try to find where the jump occurs.
