Sora, Hailuo, and Runway: generating video creatives for nutra #experiment
If university professors used to tell us to forget everything we learned in school, then this piece should start with: "Forget everything you've been told on Instagram."
All those reels where bloggers discussing AI introduce yet another "killer" tool and insist the future is already here — they all crash against harsh reality.
☝️ Let's try generating a few short video creatives for offers from the OMNI CPA catalog and see what comes out of it. Scroll down — it's going to be interesting!
A Few Important Details About This Test
As you may have guessed from the title, we'll be testing Sora, Hailuo, and Runway. Why these three? Simple — Sora AI is positioned as one of the most powerful neural networks for video generation, while Hailuo — from Chinese developers — is most often compared directly to Sora. In other words, we're working with the best results currently available. Runway made the list thanks to its ease of use and free generations.
It's also worth noting that working with any neural network requires skill. Our team has no prompt engineers — just like most media buying teams — which makes our test even more representative.
ChatGPT will play the role of prompt engineer. The task is as straightforward as it gets: we take the prompt "as is."
Generating Video Creatives in Sora AI
As you can see in the video below, hands remain the weakest point of neural networks working with graphics. A three-centimeter nail that grew out of nowhere, blister pack capsules morphing into round tablets, and other "quirks" that make using such materials impossible.
Now let's remove hands from the frame and try animating just a product packaging photo, after asking GPT to adjust the prompt for the new image. And as we can see, hands aren't the only weak spot of neural networks working with graphics.
Generating Video Creatives in Hailuo
Let's put our next offer to the test with the Chinese alternative, Hailuo. We similarly took a photo with hands (we do need to show product usage moments in creatives), asked GPT to write a prompt, and accounted for the mistakes from previous attempts. The request looked like this:
👉 Animate this photo. Move the hands and the ointment tube slightly. Minimal dynamics. Everything should be smooth with as little movement as possible. No distortion of object shapes is allowed.
We'll similarly try a different photo, and here we get our first decent result — though with almost no dynamics. There's still the question of how to cover the watermarks, but that's a topic for a separate piece.
Generating Video Creatives in Runway
Compared to more powerful solutions, Runway looks modest — yet it delivered the best result for our task. Sure, the neural network still can't handle text on packaging correctly (just like the previous options), but when it comes to animation smoothness and realism, it's a whole different story.
What happens if you upload a different photo? Let's check the second test results!
Conclusions
One final note: if you have a paid ChatGPT subscription, Sora AI will be free for you. And, as a curious coincidence, it's also the fastest. The other neural networks featured here, while offering free generations, sometimes require 1 to 3 hours of waiting before you can see the result.
Either way, everything shown above is yet another reminder that the hype around AI still far outpaces the actual results. So, while we have your attention — if you're running short on promo materials, reach out to your manager and ask for what you need. It's all in-house, so at OMNI CPA, anything is possible.
Sign up right now! And follow us on social media: Telegram & Instagram.
