Upscaling video with Artificial Intelligence

This is the first time I've upscaled an entire project to 4K with Topaz AI and I chose some pretty artifact-laden footage to use it on, from my 2011 Lumix FZ47 bridge camera. I was happy with the results about 90% of the time. The remaining 10% looked pretty plasticky so I overlaid the original footage and brought the opacity down until it looked more natural (even though it meant reintroducing some artifacts).

Here are some stills to give a better look at what the software is actually doing:

Original footage on left, upres on the right.

Original footage on left, upres on the right.

Original file with the shadows boosted to unrealistic levels.

Original file with the shadows boosted to unrealistic levels.

Same shot, processed through Topaz, with the same unrealistic shadow boost.

Same shot, processed through Topaz, with the same unrealistic shadow boost.

It’s hard not to be impressed with these results. It’s not perfect, but the algorithms are being improved all the time. This really feels like a glimpse into the future of video editing; when algorithms render resolution mostly irrelevant.