Today I’ve dabbled in some (low quality) filming + (high quality) audio recording to document my first synthesiser. Creating promotional content such as demonstration/documentation videos is an important next step, as I have my heart set on selling my electronics after graduating. The purpose is for documentation and promotion online, and I’ll be posting it to Youtube (I’ll have to make an account for my electronics) later today, as well as my personal Instagram.
Before I do so, I am interested in AI upscaling the footage, as the video was captured on a low quality digital camcorder. I like the aesthetic of the low resolution, but I think that interesting results may come from upscaling the footage. The frame rate is very low, and I think the look of interpolation with super low resolution could look fantastic. I’ve got my analogue oscilloscope wired up in the footage to provide some eye candy, and the low resolution and strange exposure makes it look very cool – with a smooth framerate, the choppy scope feed could turn into a pulsating organism.
After an hour or so of research I’m still hunting for (free) Mac AI upscaling options. ‘Flowframes’ seems to be the go-to for windows users, but is not compatible with Mac. I’m rabbitholing into open source software, and hating every minute of it.
After a further 30 minutes I am firstly feeling like lunch is needed before tackling this. Secondly I am thinking that non-ai frame interpolation might be worth a shot.
After a brief snack I’m now exporting the video (after fixing some corrupted frames and neatening up the audio a tad). Will be attempting to use normal frame interpolation rather than AI, which is a tad disappointing as I thought there might be some fun artefacts that arise from AI’s wonky ‘thinking’. Normal frame interpolation involves creating a frame between each pre-existing frame that contains the adjacent frames layed apon one another at 50% opacity. I’m a bit doubtful that this will work its magic, but will try.
A couple hours and a trip to the shops later, I’ve found a method of achieving some wacky interpolation, and am currently partway through the second attempt to render the video file: the first failed whilst I was at the shops, with no error message in sight upon my return.
The short clips below show samples of pre and post interpolation.
Second rendering attempt failed at roughly the same spot; I might just sack off the interpolation and keep it choppy. I still have ideas about using AI, downgrading the resolution and then using AI to upscale it back to peculiarity, watching what the AI does with very little to work with.