The Veo 3 effect
Trippy ads and the great social media slop unlock
Just seven weeks ago, I crowned KLING 2.0 the winner of my AI video model supertest. Since then:
Google has released Veo 3, the first model to generate synced audio (see my recent LinkedIn post).
ByteDance has released Seedance 1.0, which can create cinematic multi-shot videos with high prompt-adherence.
MiniMax has released Hailuo 02, which is ahead of Veo 3 on the Artificial Analysis Video Arena leaderboard.
It’s a rapid step change in the quality and controllability of AI video generation and it’s already having a clear impact in two areas: advertising and social media slop.
Advertising
I wrote last October about AI in video advertising creative. Two of my conclusions were:
Avoid falling between two stools. Creative should either be obviously AI-generated and you’re making a comedic/educational/stylistic virtue of that or so good that the use of AI isn’t apparent.
Fantastical/obviously-not-real scenarios provide more latitude and animation is more forgiving than attempting to simulate live action.
There have been a few ads since then making a comedic virtue out of grotesque AI video generations (e.g. Channel 4’s promo for Open House).
However, the leap forward in the quality of the latest models’ outputs means that window is rapidly closing.
Instead, we are seeing more adverts that make a stylistic virtue of the capabilities of current AI video models.
The below ad was created by AI filmmaker PJ Ace for prediction market, Kalshi, to air during the NBA Finals. He says it took him less than two days to make using Veo 3 at a cost of ~$2,000.
Contrast with this AI-generated ad created by agency Jung von Matt for German soft drinks company Spreequell:
The visuals in the Spreequell ad have more of the synthetic quality associated with the output of earlier video models.
What the ads share is a frenetic, hallucinatory quality informed by the short generations of most models, character consistency challenges and the ease of depicting the impossible (vs traditional VFX).
Despite its greater photorealism, the Kalshi ad doesn’t signpost the use of AI, relying on the beer-chugging alien to signal to viewers that the footage isn’t real (or possibly not giving a monkey’s).
Talking of monkeys, an LA dentist has been using Veo 3 to generate videos featuring a profane, beer-swilling, skydiving gorilla, and posting them on TikTok, garnering 240k+ likes.
Social media slop
There’s nothing new about inauthentic content on social media. However, the release of Veo 3 has materially increased the ease of creating photorealistic videos with matching audio.
Exhibit A: holyvlogsz, part of a trend for bringing bible stories to life using Veo 3. Here’s ‘If Moses had an iPhone…’ which has received 460k likes on Instagram:
Whilst biblical recreations are unlikely to be mistaken for authentic long-lost footage, contemporary subjects are successfully fooling viewers (e.g. the emotional support kangaroo, Ernesto, the fake America’s Got Talent contestant).
A prolific creator of AI-generated social media fodder is Paul Vu, who describes himself as a “Creator/Director that creates heartfelt AI videos to inspire positivity”. He has 23 million followers on Facebook, 12 million on YouTube, 2.3m on TikTok and 600k on Instagram. His AI-generated videos of unlikely animal encounters/rescues are generating hundreds of millions of views and hundreds of thousands of likes. This video of a buffalo rushing a school bus has been viewed 105 million times and liked over 570k times.
What’s unclear is what proportion of viewers/likers are aware these videos are AI-generated and, of those that are, what proportion care.
Like all technologies before it, AI is not inherently good or bad and will be used for both edifying and heinous ends (and everything in between).
What’s certain is that the latest advances in AI video generation are going to dramatically increase the volume of AI-generated ads and videos on social media platforms.
What next?
I suspect that the current trend of bombastic, trippy 100% AI-generated ads will give way to more nuanced use of AI in ads, as the tools evolve to allow longer scenes and easier blending of live action and AI (although 100% AI-generated will be an attractive option for companies with limited ad budgets).
As for social media, I think a tsunami of AI slop is inevitable and sadly we, the crowd, aren’t wise enough to scroll/swipe past the viewbait (content optimized purely for engagement over truth or utility), so the algorithms will continue to serve it up.
My hope is that this will be the prompt we need to more proactively curate our media diets, seeking out trusted voices amidst the cacophony of engagement-at-all-costs videos that dominate our feeds.