I don't recall the memory utilization being through the roof; I think it was CPU bound. I don't think align_image_stack is the right tool for the job.

Okay, I can manually grab the marker with G and move it to where it needs to be for the frames it loses it. After many tutorials I've finally got a stabilized movie! This one is a really good one on the stabilization process: youtube.com/watch?v=nU8zqn091rM and this one tells you how to render the stabilized video: youtube.com/watch?v=DpIXqAepfxY

I assume you mean 8 bits per color channel rather than 8 bits per pixel. For the 8-bit job that ran for 3 days I had 32 GB of RAM and an SSD.

Wouldn't each of these have a different size or amount of cropping, causing them not to align correctly?

Just tried out the blender feature tracking thing. The different light levels each day cause the tracking to stop working very quickly. I imagine the short sequences aren't going to work so well since each one is going to end up requiring a different, custom amount of cropping. The same would be true if I ran it a bunch of times using the same anchor image: the cropping would be different on all of them. I looked through the align_image_stack parameters and didn't see anything helpful.

I'd rather scale down after cropping off the edges. Besides I don't think it's going to make enough difference to allow it to complete.

Youtube supports 4k video, and I want to upload it there. Also I have a big enough monitor to appreciate the extra detail.

The movement is very slight, it can't be more than 5% of the picture. And software CAN correct for it, it's just the ones I've found that work (Photoshop) aren't designed to handle a large number of images. And having a tripod sit in the street for years is an impossibility for me.

I have 400 images and this has been running for 2 days solid and still has not written any output images. Does it have an O(N^2) running time or something?