A Bubble Water Ad That Went Viral: Entirely Generated by Veo 3}

Recently, a completely AI-produced advertisement exploded in popularity on social media, with over 300,000 views on X (formerly Twitter).

This ad was posted by the team Too Short for Modeling on LinkedIn. It’s not a commercial project but a conceptual video made for a brand they’ve long wanted to collaborate with.

Since the release of Veo 3 over a month and a half ago, AI video generation has reached new heights, especially with Veo 3’s “audio-visual synchronization” feature, pushing AI video creation into a more practical and integrated phase. It greatly lowers the barrier for video production.

Let’s see how effective this ad is.

      ^{Source: LinkedIn}

Character 1: Good afternoon, young man.

Character 2: Want to guess why I pulled you over?

Character 1: Oh, it’s not what you think.

This is “Liquid Death,” sparkling mountain water.

Character 2: Hmm. Wow. You weren’t kidding.

That’s really refreshing.

But that’s not why I pulled you over.

Oh boy.

Is it the broken taillight?

No.

Is it the license plate?

No.

Shit, man. It’s the dead guy, right?

No, sir.

Is it human trafficking?

No.

Is it the truck issue?

I don’t think so.

Then what is it?

The robbery?

No.

The bumper sticker?

No.

Is it an animal hit by a car?

Never heard of it.

Chemical waste?

No.

The roller coaster seat?

The cult thing?

Absolutely not. Hell no.

Fine, I give up.

What is it?

Because today is your birthday.

Happy birthday, Kevin.

Dad, you remembered.

Happy birthday to you, Happy birthday to you, Happy birthday to you.

The humor in this ad is dense and hilarious. But the real highlight is its remarkable “character consistency”. In just one minute, the video smoothly switches between 10 scenes, maintaining a high level of style and coherence, with the main characters and props perfectly aligned. Although there are slight jumps in details like the car windows and interior, they do not affect the overall fluidity.

It’s important to note that AI-generated videos often suffer from face distortions and object disarray.

The creative team shared their secret to maintaining consistency—super-specific prompting: providing extremely detailed, context-rich instructions to the model.

This approach aims to minimize the model’s randomness, guiding it to produce highly precise and format-specific outputs.

We previously discussed prompt optimization methods, which readers can refer to: Proven Veo3! Audio-visual sync is unbeatable, and it’s expensive for a reason

It’s worth noting that creativity, strategy, and aesthetics are still driven by humans. From initial inspiration and scripting to final editing, human input remains essential. AI is a powerful “executor,” but it operates on probabilities, not foresight—at least for now, this boundary remains clear.

Why does AI often “fail” on details?

Regarding “character consistency,” the core reason is not that models “forget” but that diffusion models—based on probability and training data—have inherent limitations:

• Models lack “understanding,” only “probability statistics”: They don’t truly understand facts like “humans have five fingers,” but learn that “five-fingered hand” is the most probable pattern. Random deviations can produce unlikely results like “six fingers.”
• Local generation vs. global coherence: They excel at detailed textures but struggle with overall structure, sometimes focusing on making local parts “look right” while ignoring logical consistency in the whole.
• From images to videos: Videos are sequences of images, and small random differences in each frame accumulate, causing inconsistencies in character appearance, background, or clothing over time.
• Training data imperfections: The data contains errors and low-quality content, which models learn and reproduce, leading to “garbage in, garbage out.”

Exploring AI’s creative potential

Today, a flood of bizarre, repetitive AI videos—like the viral “cannibalistic food characters”—reflects how AI is being “dumbed down” for quick entertainment. But the real potential lies in AI as a “creativity catalyst.”

Could these be the right ways to harness AI’s power?

• Create a parallel universe for your favorite movies.

• Imagine bringing Hatsune Miku into the world of Elden Ring—could that be the next hit game?

• Build a website for your company.

• Create a super cool concept video.

What other surprises do you think AI can bring to our creativity? Share your ideas in the comments!

]