Last Update: 19 Aug 2026
This comes up enough on my youtube.com/@stevetures page that I've decided to write down a page of ehtics, as I regularly (and reasonably, in some cases) find my work criticized as slop. Here's the code of ethics I've chosen for myself to adhere to.
I have never and will never upscale using dirty power or data centers.
I will never monetize my page. This is my hobby, and happily I have a day job.
I publish only for claimed work**. In other words, for copyrighted works, if the copyright is not claimed for some reason on Youtube and there's an official upload (i.e. the claims mechanism broke) I'll delete it and never publish it publicly.
I will unpublish any work that does an official 4k version (unless that 4k version is notably bad, but I've yet to see this happen).
I will never upload something that doesn't pass my own bar for quality. In other words, I have and will continue to fail attempts that simply have distracting upscale artifacts.
For example, I've attempted the Nirvana - "Come As You Are" and with all prevous and current attempts, the models struggle to correctly interpolate the faces of the band and it always creates blobby faces. While it's not a perfect bar, I shoot for something that could pass the average viewer's notice for how distracting the content is.
I will never use tools that train on copyrighted data, though this is an easy one as that's not how upscaling works anyways.
** - There's a few videos that are extremely obscure and don't even have official uploads from the music band or owners, so Youtube can't claim this. I encourage the rightful owners to upload their version and to properly claim (and hopefully allow) my version.
Energy
My work is done at my house, which is happily solar-powered using my own solar panels (i.e. not rented), and the work is done solely on my workstation, which contains a single slightly overclocked nvidia 4080 card and an i7-1300k CPU to support this work.
My work is never rendered in the cloud aka not at a data center.
Tools
I currently use Topaz Video, though I'm investigating other open-source tools.
How does it work?
AI upscaling works differently than Generative Image and Video work, in that the input stream is specific and structured (a time-related grid of hundreds of thousands of pixels per frame, times ~30 frames per second) and not a text question or request, and the models are trained less on existing artwork (which I would argue is art style theft) and more towards filling in the missing pixels between existing ones for any given small grid of pixels.
The models have definitely outputted some cringe in the past, as there's limitations to this work.
Who am I?
I'm a hobbyist. On the side, I have a filmmaking degree, and an audio engineering certification, and I'm a multi-instrumentalist and I'm working on some documentaries, one of which I'll share with the world soon. I'm also an outdoor and geology and earth sciences enthusiast.
I do work in software and digital realms of art and communication as a day job, and I accidentally helped to invent some more accessible versions of video calling to enable everyday humans (and not just CEOs) to sit in a room for cheap and talk with one another. Prior to that, I had a career in IT and helping people with their computer questions and requests.