I'm new user of Kdenlive (v. 22.08.1, flatpack, debian) and as far as I read and experiment I cannot find a workflow for the following situation:

- There is a source UHD jerky clip. The object of interest is small, somewhere in the center of the frame but the whole frame is very unstable.
- The target output is a stabilized HD video, cropped frame of the source, 200% zoomed, in order to minimize detail loss.

Motion tracker works perfectly and its produced keyframe data are applied to the transform effect (copy-paste).

The problems start when the transform zoom comes into play:
- If the motion tracking keyframes get applied to a subsequent zoomed 200% transform, they don't correspond to the original movement (which is reasonable, because they are absolute 2D coordinates).
- If the zoom transformation effect is applied first and the motion tracking second, on the zoomed frames as they appear on the project monitor, the tracking analysis seems to ignore the zoom and gives results as if it was based on the original (not-zoomed) clip.

I managed to get the desired result only by doing the motion tracking on a UHD project, exporting the keyframe data, and switching to another HD project to apply the zoom and tracking keyframe data. But I hope there should be a simpler method to achieve the same.

Questions:
1. Can the motion tracker take into account the previous geometry altering effects in the effect stack?
2. Is there a another/better method/workflow to stabilize and zoom-crop a clip?

Thanks for your time and attention,
Thanks to the devs and the community of this excellent software.

Motion tracking is not working well on zoomed/croped videos. also on "transformed" video... when i have to do something like that i use the motion tracker on the video without effects .. i use generated "points" to do what i need (a pointer.. .an inversed transform for stabilization or something else ) after that i render the clip.. and after that i use "stabilization tools" or zoom... or crop on the new rendered clip (importing it) ... any other way right now can do random things... (on timeline .. or only on teh rendering)

Thanks for your prompt response.

bartoloni wrote:.. i use generated "points" to do what i need (...) after that i render the clip..

After experimentation I got to a simple solution, without intermediate rendering. The steps are:

Run the Motion Tracker to get the keyframes
Apply the keyframes to the Transform effect to have a stabilized video (with dancing black frame around)
Add a new instance of the Transform effect to the stack and adjust the zoom (and potentially any other parameters, keyframable)

The result is pixel perfect for a 2160p clip, zoomed 200%, and rendered to 1080p! (tested with resolution patterns).
As far as I understand the multiple transformation stages do not alter the resolution of the video, but instead add up the calculations that will be used for the final output. This is a problem-solving tool!

Additionally, while in the import keyframes dialog, the project monitor shows the result of any geometry and positioning parameter change, for any frame of the clip. Fantastic!

Thank you for your time and attention.