The best is still to use Mocha or Nuke. They need a lot of work but you gonna get as much as you can from stabilization. Mocha is probably more straightforward and less processor intensive than nuke which requires both the expertise and the hardware. They also both allow to save more image by reconstructing the "out of the frames" from different frames which comes quite handy especially knowing cropping is a major issue with stabilization. Again it's not magic, it's always better to have a very solid shot in first place, don't expect miracles.

I use the free & open source vid.stabfilter from Georg Martius, I apply it using ffmpeg's "vidstabdetect" and "vidstabdtransform" video filters.

See also this posting for example video links.

The procedure I follow is:

(1) Cut the relevant parts out of the camera recorded files in the compressed domain (not re-encoding, only cutting at I-frames). This is relevant (apart from saving a lot of time and storage) to not confuse the software stabilizer by irrelevant and possibly wildly moving scenery outside of the range I actually want to stabilize. I do this using mpv and the excerpt plugin.

(2) I invoke a script that calls ffmpeg (which has been compiled with vid.stab support, you need to install libvidstab first and tell ffmpeg to use it when configuring before "make"). I've written two variants of this script, one for video with intentional motion in it, and another for video that should look as if the camera was mounted fixed on a tripod. The first script looks like this: ("\" at the end of the line is a line continuation for "bash"):

 ffmpeg -i input_name \
  -vf "vidstabdetect=result=name.trf:shakiness=5:accuracy=15:stepsize=6:mincontrast=0.3:tripod=0" \
  -c:v rawvideo -f rawvideo -y /dev/null

 ffmpeg -i input_name \
  -vf "vidstabtransform=input=name.trf:smoothing=10:optzoom=1:interpol=bicubic:tripod=0" \
  -c:v libx264 -profile:v high -preset:v faster -tune film -b:v 100000000 \
  -c:a copy -c:s copy -c:d copy -map 0 \
  output_name.mkv 

The second version of the script (for the "fixed tripod"-like scenes) looks similar:

 ffmpeg -i input_name \
  -vf "vidstabdetect=result=name.trf:shakiness=3:accuracy=15:stepsize=6:mincontrast=0.3:tripod=1" \
  -c:v rawvideo -f rawvideo -y /dev/null

 ffmpeg -i input_name \
  -vf "vidstabtransform=input=name.trf:smoothing=10:optzoom=1:interpol=bicubic:tripod=1" \
  -c:v libx264 -profile:v high -preset:v faster -tune film -b:v 100000000 \
  -c:a copy -c:s copy -c:d copy -map 0 \
  output_name.mkv 

(3) I import the resulting video file into my NLE (kdenlive in my case).

(The choice of constant bitrate 100MBit/s for the encoding is meant to retain about the quality of the original 100MBit/s files recorded by my LX100, you could of course prefer other formats/parameters for your intermediate files.)

I just noticed that my stabilized footage has more contrast than the original. I am wondering if it is because my gh4 range is set 0-255. I am definitely losing delicate highlight detail. This is confirmed on scopes. I am using mercalli v4.

Because stand alone is more versatile for import / export since it plays the role of third party software and can adapt to any workflow. Plugin only works for 1 software at a time.

@Jspatz

I just noticed that my stabilized footage has more contrast than the original. I am wondering if it is because my gh4 range is set 0-255. I am definitely losing delicate highlight detail. This is confirmed on scopes. I am using mercalli v4.

My Gh4 0-255 Mercalli v4 stabilized files looks just the same as the originals. 0-255 levels both. The previous version converted 0-255 to 16-235.

Is your editor doing some converting or clipping?