Using Ai To Convert Classic Cartoons Into 16:9
Hypothesis
If you're a fan of classic cartoons, you've probably noticed that they don't quite fit in a modern TV. That's because most classic cartoons were originally created in a 4:3 aspect ratio, which is narrower than modern TVs. This can lead to black bars on the sides of the screen, or much worse distorted/cropped images, as service providers try to remove the empty space via distorting the image, or scaling it up to fit (see Disney Plus Simpsons.)
This got me thinking about the using some of the latest Ai advances to see if you could artificially expand the frame, without the need to distort or scale the image, effectively creating new data that matches the old 1:1.
Below are some Ai Outpainting tests I ran on various classic 4:3 animated series. To my surprise it worked pretty well, but not without some major caveats, which I will go over at the end.
Batman: The Animated Series
Spider-Man: 1967
Classic Looney Tunes
Conclusion
Yes, it is very possible to extend the static background paintings with current Ai, it excels at matching the style of the input image, and can turn around multiple options in seconds. However no Ai I can find is temporally stable yet, or understands character/FX animation to the degree of a human artist, and to be honest, I hope that never happens.
Pros
Can easily extend static BG paintings that don’t have camera moves or overlapping character animation at the edges of the frames
Fast, able to generate multiple extensions in seconds, and as many as you want
Is somewhat guidable with the use of text prompts, but results are mixed
Cons
No temporal stability, only works on stills out of the box
Changes the original creators intent (but less destructively than distorting/scaling imo)
Any animated camera moves would require a human to stich together the full BG painting and then extened via Ai, and then back to a human to track/copy the original camera animation for the extension
Any characters or other animated elements that overlapped with the edge of the frame would require a human to extend the animation into that region
Struggles with really abstract elements that don’t have an easy to define analog in the real world
In these tests, I am using Adobe’s “Ethically“ trained Ai, but there are still many questions and concerns about this technology, and although I don’t think something like this would ever be cost effective with today’s non-Ai options, there are still valid concerns about pursuing its use in all artistic endeavors.
All these tests were done using the Photoshop Beta Ai v24.6
