Inpainting - The Simpsons

Looking back on some of this stuff it's crazy how janky and low quality the generations look- especially compared to the state of the art today. This was originally stuff I did around 2020, and as an update on this post in 2026 it's incredibly stark the difference. With seedance, kling making strides in video gen, and midjourney carving out the og image gen space this stuff looks primordial.

At the time I thought it was pretty amazing though. And perhaps because it wasn't really *me making the art, I have a level of removal that I don't sometimes seeing my own artworks that I or my parents kept around. That cringe at whoa how did I ever think that was good or worth making. Seeing the models or yesteryear feels akin to seeing the cave paintings of ancient civilizations. The early beginnings of an intelligence? Creative lifeform? I don't know if I see ai as something that comparable to humans but it's probably as close as I feel to animals. Maybe growing up with disney and pixar and the normalization of anthropormphizing has made it feel natural. I think I tend to think of ai as closer to us than to some god-like figure. Perhaps it will surpass humans in power, prowess, ability- I think in many ways it already has.

Does that mean we aren't building god as a product or creation, but as ourselves? I don't know. There's a level of non-intention that I associate with the current climate. Some refer to the prisoner's dilemma or some kind of arm's race. Not sure if those are the right analogies, but I also agree there is a level of maybe we're bring about something and it's not completely by choice.

Over the years, there have been many depictions of god or gods or god like figures. Perhaps the one we are closer to is just the initiation of something- the big bang? The catalyst to a reaction?

I was initially drawn to this project because I was just so impressed with the idea of using a machine to create images. Think dall-e and some specific papers, motionflow? I forgot the names, they were pretty niche. I just wanted to explore and try implementing the tools myself. Was looking for a usecase. I was reseraching and apparently millions of dollers have been spent to 'remaster' or retrofit? or somehow make an existing work better formatted for a different screen or resolution. There was apparently an uproar of when simpsons came to disney plus. Where a lot of the older episodes were created with the square box format, but I guess a lot of people find it offputting? Somehting about having black bars on the sides of a widescreen tv seeming like an error? So as a way to get around it they just zoomed in. Which at first glance solves the issue but the problem is it cuts off a lot of information. Simpsons isn't super serious, it's mostly humour and some say you can get a lot of it just from the audio almost like radio. But there are some scenes where the jokes are highly contextual and cutting off up to 30% or more of the scene actually ruins the joke.

A few examples:

Plenty of media outlets noted the 'backlash'

I did some napkin math and thought if I could figure out how to remaster the series with these new ai-assisted methods, might be able to get disney or somebody to pay me hundreds of thousands as opposed to millions. A 10x reduction.

I toiled away at it for weeks, months and it became obvious to me that it just wasn't ready. A lot of hallucinations, artifacts, weird warping. Simple black bars or using clone stamp from traditional photoshop seemed better option to me.

I tried some tests with old movies like casablanac and those seemed even worse as the bar for seeing something off is much higher for film I think as opposed to cartoons. But was fun to experiment with the early beginnings of generative ai.