Originally requested here: https://community.topazlabs.com/t/video-enhance-ai-feature-requests/20542/140?u=shikuteshi
The concept is simply instead of outputting images in case of an issue that causes VEAI to stop and lose progress if one were outputting an mp4. VEAI would instead output mp4’s into sections. (configurable).
Lets say I have a 100,000 frame video. Every 1000 frames, or whatever I set, would have VEAI save that as a new video file. It will do this for all 100,000 frames until I end up with 100 video files. Once VEAI has recognized that it has completed processing all the frames, it would automatically stitch the video files together.
This solves a few things.
- If someone chose to output a video files instead of an image set, it would potentially save them from hours or days of progress in the event of some unforeseen issue like a power outage.
- Because a CRF value can be set on video you would save a lot of disk space in comparison to PNG’s. (jpgs would save more space than PNG’s but I would never recommend it)
- If VEAI could support the use of ffmpeg that we could point to our own copy via settings then we could save a lot of post processing by having the initial file format already be in the format we want. In my case I would’ve loved in VEAI could have HEVC 10-bit output already done for me so that I don’t have to spend a few hours afterwards encoding.
And I want to expand on the original post.
There’s an open source program known as Av1an that basically does exactly what I’m looking for but in a different way.
Av1an is a video encoder that supports AV1, VP9, H265, and H264 encoding but in a unique way. It uses scene detection to determine how to section parts of a video off into smaller parts. Then assigns each video to it’s own dedicated worker program with a user set amount of threads.
It’s very efficient since it scales nearly infinitely unlike your typical encoder. But that’s besides the point.
It does this quite well. The determining of how to split apart a video and then merging them back together. And because the videos are in parts, it can encode, stop, and resume without much loss of progress. So we know that attempting something like this is perfectly possible, and workflow wise is very saving a user of lost time in the effect of an accident.
I implore the VEAI team to take a look at Av1an and come up with a solution to providing something similar. Video outputs are way better than image outputs due to file size alone. But you run a risk the longer the video you’re trying to work with.