Although it is true that AI cannot *recover* details, it can estimate what those details would have been, and it ends up creating video with high detail and high fidelity. Some services may compress these files when uploaded (i.e. the primary one of this sub, redgifs) but the quality loss from that is negligble for most use cases. If you wanted ultra-high-quality video, then yes you’d have to upload to a file store rather than a media store since file stores do not use lossy compression on content.

I use H.265 (HEVC) encoding on my content so that the file sizes can remain small with high fidelity and detailed video footage, for an optimal viewing experience. My typical workflow is usually: Video -> Enhancement 1 (Remove Artifacting, Add Granular Noise) -> Enhancement 2 (Increase fidelity and detail) -> Frame Interpolation (24, 30 -> 60/120/144 FPS). FPS typically doesn’t matter since the videos compressed and uploaded to Redgifs are scaled to 1080p60 anyways, but it can make slight differences in the apparent smoothness of video, which I enjoy looking at. The higher input quality prior to the forced compression, the better it will look. Ideally you want to get as much as you can out of a given bitrate as possible, as with many videos.

The software I use is Topaz Video AI, which has a $300 fee for a yearly license. It’s quite expensive but I like doing this for fun. It’s considered the state-of-the-art AI upscaling software. I am happy to take requests to upscale for free (I do not animate myself.) I will do them for free when I have time, and I do not feel comfortable asking for payment as my work isn’t really difficult and much of it is done with existing art created by talented artists, who you should check out yourself.