The big picture: YouTubers are using neural networks and other software tools to repair, upscale, and colorize old video footage, so that it can be experienced in a different way on modern hardware. However, academics believe that could open a can of worms as the original footage is changed to a degree that removes the context of how it was captured.
Earlier this year, we saw how a YouTuber was able to achieve remarkable results in upscaling old, low-resolution footage from 1896 to 4K video at 60 frames per second. This raised the question if “Enhance” would soon become more than a television trope, and whether AI tech could be used with all the old video and audio recordings that are public domain.
Since then, Denis Shiryaev’s channel has exploded in popularity as he’s kept adding more upscaled videos and colorized black and white footage that attempts to recreate the various nuances of the 19th and 20th century. But this is merely a showcase for a larger project, a Poland-based company called Neural Love whose mission is to use neural networks and algorithms to clean up, upscale, stabilize, and colorize historic images and video.
For many people, the resulting content has captured their attention and transported them back through time. However, it turns out not everyone likes the new trend. Some historians of art believe these modernisation efforts are problematic and offer a distorted image of what was actually captured in the original content.
Emily Mark-FitzGerald, who is Associate Professor at University College Dublin’s School of Art History and Cultural Policy, notes that “The problem with colourisation is it leads people to just think about photographs as a kind of uncomplicated window onto the past, and that’s not what photographs are.”
Neural Love’s Elizabeth Peck told Wired the company does explain to its clients the way it approaches “the restoration aspect and the enhancement aspect.” For instance, the removal of noise and imperfections is seen by the company as merely the process of “returning the film to its original state.”
Academics disagree, but they worry more about how colorization and frame rate smoothing in particular add in details that were never captured on the original film. For Luke McKernan, lead curator of news and moving images at the British Library, these techniques turn out subjective interpretations of the original content, which isn’t acceptable as it removes the authenticity and the accuracy of the source footage.
McKernan argues that “colourisation does not bring us closer to the past; it increases the gap between now and then. It does not enable immediacy; it creates difference.”
In any case, Neural Love knows the resulting videos are not historically accurate, and they offer the option for clients to add a watermark so that they won’t be taken at face value. Whether or not that’s enough, only time will tell — but people scrolling on, say, a Twitter feed might not take the time to search for the original or even realize it is not meant to be historically accurate.
A somewhat similar debate could spark around video game remasters. Take the recently released Crysis Remastered — it comes with 8K textures, software ray tracing, voxel global illumination and a host of other changes. However, for many it lacks the kind of attention to detail that made the original a great game that’s still talked about 13 years later. It still melts video cards, but it was never just about that — Crysis was a seminal point in the history of video game graphics, and the original is still the definitive version that will offer you a window into that era.