How hard was it to create a fake soda ad with AI?
I used Sora to create a ridiculous ad for fish-flavored soda. The results were...interesting.
Disclosure: This post was written by me, a human, with no AI assistance. Excuse the typos!
In a previous professional life, I regularly worked on complex video projects with detailed storyboards, multiple artists, and six-figure budgets. My role typically involved developing the overall concept and writing detailed scripts before handing everything off to art directors and videographers for storyboarding and production. It's a finicky, complex process filled with trial and error…which is exactly why I'm deeply skeptical of AI-powered video makers.
And yet, the idea is intriguing. YouTube is already filling up with AI-generated video—hello, Italian brainrot! And big brands like Coca-Cola are running AI-generated ads. Could I create something similar, just for fun?
Since my kids and their cousins recently invented a hilariously absurd soda brand—CanZo, a fish-flavored soda with floating chunks of mystery meat—I figured now was the perfect time to put these video tools to the test.
For each platform I try, I'll attempt to create a commercial for CanZo, armed only with an AI-generated product design and one seriously raised eyebrow. This post will look at Sora, OpenAI’s text-to-video model.
Sora: One-click video from text and images
Sora is designed to generate realistic, high-quality video clips from written prompts. Unlike older AI video tools that rely on templates or stock assets, Sora builds entire scenes—including motion, lighting, and camera angles—from scratch.
You start by either generating or uploading an image, and you can define “presets,” which specify camera type, angles, and other frequently used image specs. Then, once you have an image you like, you can prompt Sora to build a video clip.
As a user of ChatGPT Plus, OpenAI’s least expensive subscription tier, I can only generate five seconds of video at a time with Sora. But I had high hopes that it could create a usable snippet for my fake soda ad.
The image
For this test, I uploaded an AI-generated image of a CanZo soda can based on a prompt devised by 8-to-13-year-old children.
The prompt
Once I had uploaded the can image, I instructed Sora to create a short video clip:
CanZo is a new breed of soda, a fish-forward carbonated drink with a salty finish and small chunks of mystery meat. Please show a can of CanZo (image provided) sitting on a white sand beach. The sky is blue and inviting. Palm trees sway in the background. A foamy wave crashes over the can and then ebbs away. CanZo is still standing.
Round 1: Not enough sand
The first clip was a bit underwhelming. It began with the can floating against a drab background. Then, highly animated water flowed behind it and disappeared, changing the background to a drab blue. No sand or palm trees were present in any form.
Round 2: Sand and a mutating brand
Not satisfied with the first attempt, I used Sora’s remix feature to basically ask for a do-over. I emphasized the importance of sand and a beachy feel. And the tool did take direction, replacing the bland background with fairly realistic sand.
But some big problems remained. The can’s design had totally changed along with the tagline and verbiage. The crashing wave visibly mutated the can. And the sky in the final scene did not look realistic.
Back to the remix tool I went.
Round 3: An uncanny can with tropical vibes
I asked for one final round of updates, focusing on the importance of placing the can on a realistic tropical beach and reverting to the original can design. And, despite the awkward water transformation and continued loss of can branding, it was Sora’s best attempt so far.
The verdict: It won’t be replacing video pros any time soon
While this test was fun, it also underscored the vast distance between the AI hype (one-click videos!) and the reality (video that needs a ton of human editing to be usable). I suspect I could get better results by tinkering with presets and learning from more experienced Sora creators.
I am also interested in seeing how Sora compares with newer text-to-video platforms. (And my kids are still hoping for a CanZo ad they can post on TikTok!)
Until the next experiment…