Bursting the Bubble on 'Effortless' AI

It’s only 81 seconds long, but Balloon Head, produced by the experimental Canadian outfit, Shy Kids, quickly became a poster child for what the brave future of AI-powered filmmaking would look like.

The film, which you should watch here, was lauded for the way in which it structured a series of clips, all produced by Open AI’s yet-to-be-released 'Sora', into a cohesive narrative. The VO and narrative was all penned by humans, so the showcase was really about how physical, believable and consistent the AI generated footage felt.

At Mr President, we’ve spent the last year building out our AI tool portfolio - with regular workshops covering everything from Large Language Model etiquette to the ethics of image generation. It’s been fun, upbeat, and experimental, and has already helped us seriously update our ways of working. But when we sat down to watch Balloon Head, as part of a session all about “AI’s cutting edge”, the reaction from our creative team was, to put it mildly, lacking in enthusiasm. 

“We’re all fucked.”

The thought of high quality video assets, instantly created with almost no effort, felt with like something to fear. But they needn’t have worried so much, because, as Patrick Cederberg, Creative Director at Shy Kids, has revealed in his interview with FX Guide, creating video content with AI is anything but effortless and it certainly isn’t instant.

In the conversation, Patrick says that for every clip that was usable, “[there was] probably 300:1 in terms of the amount of source material to what ended up in the final.” 

Even with some bad maths, and a conservative estimate of sixteen clips in the final video, that still puts Shy Kids at almost five thousand video generations to review. That’s around twenty hours of AI generated footage to watch. And that’s before any of the serious post-production that was needed in terms of upscaling, colour alterations, and removing the spooky balloon face that Sora seemed so incessant on adding to footage.

The fact is that nothing about content production with AI is genuinely effortless. At least, not if you want the end result to be of a passably acceptable quality. It just requires a new type of effort. One that places more focus on reviewing, editing and synthesising, instead of creation, craft, and production.

Finding the single minute of usable footage within the twenty hours of haystack is a remarkable feat. Ignoring all the hallucinations and distractions that Sora must have churned out, and wilfully discarding almost everything generated to find the ideal footage, necessitates a remarkable focus and a strong, unwavering sense of artistic direction. Not to mention the more technical skills required to effectively prompt Sora, a new and mostly untested tool, in a way that will produce the right sort of content.

The promise of absolute automation that AI tools offer is seductive. But it’s just re-selling the dated 1950s dream of a robot army picking up all the chores, whilst us humans kick back and relax.

The reality is that AI, at least in its current state, is just another tool for us to use. A tool that comes with its own skill requirements and workflows. In the next few years, only those teams with the technical skill, the genuine creative vision, and the capacity to wade through the bottomless cutting room flow will be able to wield AI to produce impressive results.

That’s why, at Mr P, we’re spending so much time upskilling ourselves on how these tools work. It’s why we’re doubling down on human-led creativity, and it’s why we’re all planning on getting a little more used to the editing suite.