AI video platforms aim to overhaul creative workflows

Video produced by AI was once a novelty—sometimes impressive, but more often populated with off-putting human distortions and spectral graphics. But in the last year or so, rapid technological progress and new professional workflow tools have made video AI much more useful, experts told us—and finally able to yield output good enough for, say, a Carl’s Jr. TV spot.

At creative agency AKQA, AI now plays a role in most early-stage creative experimentation, a relatively new development within the last eight months, according to AKQA CTO Ben Royce.

In fact, Royce said the quality of video AI tools reached an inflection point this year. Output long ago graduated from the extra fingers and distorted faces of early offerings, and Royce said the software now seems to better understand the jargon and needs of video editors (AKQA mostly uses an internal tool from parent WPP that’s powered by various video models).

“The tools are finally at the bare minimum to start working with reliably,” he told Morning Brew.

So what does the agency bring to the table in a world where creative AI models have reached that level of sophistication?

“[Clients] are paying for the skill and the talent to be able to use [AI] effectively at their brand’s quality or higher, and they’re paying for taste,” Royce said. “It’s easy to produce slop…The difference is we know how to structure, create lots of assets that are good enough on the first pass or require minimal editing. That’s hard, right? That’s actually a craft.”

Although OpenAI scrapped its video platform Sora to free up resources earlier this year, progress in the field of video AI has continued apace. Other experts agreed that it turned a corner in the past year, both in terms of the usefulness of the tools and thorny legal issues.

Many of these tools focus on creative production workflows with multi-model platforms that cater to marketers’ needs. As of last year, nearly nine in 10 digital video buyers said they used or planned to use AI in video creation, per an Interactive Advertising Bureau survey.

“When I watch TV today, there are so many commercials that are made with AI,” Caroline Ingeborn, COO at video AI model startup Luma, told Morning Brew. “They don’t tell us that it’s made with AI, but I know it’s made with AI because I’ve watched a lot of these videos. So this is already happening.”

Winning over creatives

Ingeborn said Luma, which has worked with agency holding companies Dentsu and Publicis, has an immersive process for onboarding agencies and marketing departments to its video tools. Over the course of two weeks, Luma works side by side with creative staffers on multiple concepts, which might include mock-ups for pitches or localization opportunities, she said.

“That helps with one question that I often get from the executives, which is, ‘This sounds awesome, but I have 500 creatives working for me, and they are shit-scared of AI. They don’t want to touch this. How do I tell them to touch this?’ You get another creative professional to come in and show them how their everyday can be much, much better.”

Ingeborn said Luma tries to frame the platform to creatives as a way to produce ambitious ideas that they wouldn’t otherwise be able to realize, rather than a cost-cutting measure.

For AKQA, one of the biggest AI timesavers is uploading real 3D graphic assets from a brand like Lexus, then using AI to produce “hundreds, if not thousands” of variants—the Lexus SUV driving through the snow in Sweden, the car driving through the streets of Barcelona or Brooklyn in different weather and lighting, Royce said.

“All of a sudden you can do a lot more testing because the barrier to entry to more variants is not flights and hotels, it’s creativity,” Royce said.

New creative AI platforms are also helping professionals toggle between many different video models, allowing them to pick and choose between AI offerings depending on which is better suited to a given task. Joaquín Cuenca Abela is the co-founder and CEO of one such tool, Magnific (formerly Freepik), which has been used in campaigns for Carl’s Jr., Puma, and Amazon Prime’s House of David.

The company started its pivot toward AI relatively early in the generative AI revolution, as Cuenca Abela saw what nascent models like OpenAI’s Dall-E were capable of. Rather than compete with Big Tech on engineering models, Magnific wants to give creators access to a wide range of top models.

Legal questions

Cuenca Abela told us that brands were initially wary of the legal and copyright gray areas surrounding video models—at least for any external marketing. That’s started to change as various rulings have come down from the courts around fair use. In some cases, the sources of video training data have also become more publicized, he said.

“Today, all the big video models are backed by a social network—that’s where they are getting their video,” Cuenca Abela said. “So it’s a combination of more clear provenance of the data used to train the models and some rulings that are giving [brands] a peace of mind that this is actually OK.”

AKQA’s Royce tells brand clients to only use enterprise platforms that won’t poach their intellectual property to train future iterations of models. He said model providers have become better about ensuring legal safety overall.

“The industry has now come to a point where finally it’s mature enough, where model providers took client concerns seriously,” Royce said. “They provided indemnity or proof that they weren’t going to either retrain on their data or use other people’s data that they hadn’t [gotten permission to use].”

Marketers are also worried about the quality of the video in which their valuable brand assets will appear, according to Hannah Elsakr, VP of GenAI new business ventures at Adobe.

Brands are interested in AI only “when it’s protective of their IP and they see that the output—at a bit-depth, at a quality, the color, the character, the physics—is in keeping with their brand,” Elsakr told Morning Brew.

No matter how sophisticated AI video tools get, however, AKQA’s Royce said there will always be a place for talented creatives who know how to get the best material out of these models.

“You can give the best tools to someone who doesn’t know what they’re doing, and it’s not gonna end well,” he said. “There’s a craft and a new skill set of knowing how AI models respond to things and what they’re good at, and what they need to get really good at, and that’s something that I think the entire industry is gonna have to pick up.”