I am processing around 20 videos shot with a consumer camcorder in 2002. Files are MP4, SD quality, averaging 15 minutes each. Output target is FHD (1920x1080) for YouTube upload.
I am currently using Artemis LQ + Recover Detail 35 + Denoise enabled, which gives a noticeable improvement but takes approximately 1 hour per 3-minute clip on a RTX 2080 Ti.
I accidentally processed one clip without any enhancement settings and it took only 3 minutes with already some visible improvement.
My questions:
What combination of model and settings would you recommend for this type of footage?
Is there a good compromise between quality and processing speed for SD camcorder material?
Is Artemis LQ really the best model for this use case, or would Proteus be faster with similar results?
Artemis LQ is a good option depending on the condition of the video files.
You can also try the preset MiniDV HD Int Basic as an option, and swap the video format to Interlaced Progressive if the source has already been converted from an Interlaced format if it was original stored in one. This is a custom preset that the Product team worked hard on finding settings for that worked with a wide range of tape and older content that needed enhancement as well as upscaling.
Thank you for your answer. I will try the MiniDV HD Int Basic preset.
Two questions:
Is this preset faster than Artemis LQ? Currently processing takes about 1 hour for a 3-minute clip on a RTX 2080 Ti, which seems very slow.
My files are MP4 exported from consumer camcorder footage originally shot in 2002. Should I consider them as already converted from Interlaced, and therefore use the Interlaced Progressive option?
Processing will probably be a little slower as the Proteus model with manual settings has a bit more going on. To increase the processing speed you would need to upgrade your GPU card to one that has more VRAM and faster throughput. The RTX 2080 Ti is an older card and is going to struggle with some of the models when upscaling video.
I’d suggest that you also try Iris for that kind of source (and there I still prefer the old LQ V1 model as the newer Iris variants tend to create strange stuff every so often).