Ever find yourself squinting at a video, trying to figure out if it’s real or AI-generated? It’s getting tougher, and I just saw a video that proves we’re entering a whole new era.
An AI professional I follow just got early access to Google’s new AI video generator, Veo, and the results are pretty mind-blowing. He put it through a massive test, and I had to share what he found.
⚙️ How Google’s Veo Works
The creator explains that once you’re in, you have two main options:
- Text-to-Video: You write a prompt, and it creates a video. The expert recommends getting detailed with your prompts for the best results.
- Image-to-Video: This mode first generates an image based on your text, and then you add a motion prompt to animate it. The YouTuber noted that you can’t upload your own images for this yet.
For either option, you hit “create,” and in about 3-5 minutes, Veo gives you four different video clips to choose from. This is a game-changer because you can cherry-pick the best one!
🚀 The Ultimate Stress Test
This is the best part. The mind behind this video didn’t just test a few simple prompts. He asked his followers on social media for ideas and ended up testing over 30 of their wildest prompts. He committed to showing the first four results for each one: no cherry-picking, just the raw output.
He tested everything you can imagine. Here are a few highlights:
- ✅ “A dog surfing in La Hoya”: This one came out surprisingly realistic. One of the four generations looked like a real drone shot hovering over the dog.
- ✅ “Bollywood dance sequence in Alaska featuring Grizzlies and polar bears”: I was shocked by this one. The creator showed how well Veo handled this complex scene. The results were incredibly detailed and well-composed.
- ✅ “A punk rock squirrel…shouted into a microphone”: This was just fun, and the videos were pretty solid and hilarious.
- ❌ “A man is playing with his dog at the beach in the background dolphins are surfing”: The YouTuber showed that Veo struggled here, just like other models. It had trouble combining two separate, complex actions in the foreground and background.
- ❌ “Will Smith eating spaghetti underwater”: Nope! The expert demonstrated that Veo has guardrails that block you from using real people’s names.
✨ Veo vs. Sora: The Showdown
For every single prompt he tested in Veo, the creator also ran it through Sora to see how they compared. His verdict? Right now, Veo is consistently outperforming the currently available version of Sora (Sora Turbo).
While he acknowledges that OpenAI probably has a more powerful Sora model in the lab, he concludes that of the tools people can actually get their hands on, Veo is the new king.
This is an incredible leap forward for AI video. As the creator says:
This is the worst this technology will ever be, and it’s already this good.
I’m blown away.
For the full deep-dive and to see all the wild video generations side-by-side, make sure to watch the original video from this innovator, as he shows everything!