AI use case: Making movies!

Playing with Veo 3 is good fun, but is it the end of the world as we know it?

It took a little longer than expected, but we finally got access to Veo 3 here in Canada—and, of course, I was dying to get my hands on it.

There are limitations, naturally. I had the impression from the initial marketing that Veo 3 would generate videos in 4K, but in reality, the clips are rendered at 720p. You can upscale them to 1080p, but that’s a far cry from true 4K. There’s currently no support for vertical aspect ratios—no way to force portrait orientation—and crucial features like Extend and Ingredients are still only available in Veo 2. But once you've experienced the fidelity and flow of Veo 3, you’ll find it hard to go back to its predecessor. It feels like a step into the future, making Veo 2 feel outdated by comparison.

At the moment, we’re limited to 8-second clips, which feels constraining—but understandable given the compute load. Censorship is heavy-handed with Veo 3, and the service isn’t cheap. Plus, it fails fairly often: in my estimation, around 30 to 40% of the outputs are completely unusable—glitched, irrelevant, or visually incoherent. That said, the remaining 60 to 70% range from startlingly acceptable to occasionally excellent. When it works, it really works.

The model’s greatest strength is in dialogue-driven cinematic moments. Veo 3 shines in quiet, composed shots where the emotional tone is clear and the action minimal—characters talking, reacting, reflecting. It can do a passable job at mid-paced drama scenes. But when it comes to fast motion or action sequences, the model stumbles badly. Motion blur, unnatural physics, and bizarre transitions quickly degrade the illusion. At its worst, it’s unusable. Still, even with these flaws, Veo 3’s capabilities are far and above what any other publicly available video model can do.

The closest competitors I’m aware of are Higgsfield and Hedra—both impressive in their own ways, but neither has quite reached the same level of cinematic coherence or aesthetic control. Veo 3 sets a new standard, even if it’s still heavily shackled by training boundaries, safety filters, and resolution constraints.

Nonetheless, Veo 3 is a joy to experiment with. Even limited to 8 seconds, even at 720p, even when you're tiptoeing around the censors—it still feels like you’re holding a piece of the future in your hands. Can you imagine what this will be like when we get access to longer durations, aspect ratio control, better fine-tuning, and less red tape?

It’s one thing to know that it’s now possible to generate realistic fake videos from a text prompt. But the feeling of actually doing it—typing a sentence, waiting a few seconds, and watching a photorealistic, dynamic video emerge from nothing—hits differently. Once you feel that firsthand, it changes you. The news, the media, fiction, politics—it’ll never look quite the same again.

In the meantime, I invite you to enjoy Chrome and Consequences, my short fake documentary about the robot uprising of 2033. Yes, I’m fearmongering for fun. No, I don’t actually believe in a looming robot apocalypse—but I couldn’t resist. At heart, I guess I still love making disaster films. Veo 3 just gave me a shiny new excuse.

---

Creator’s Notes:

There was very little cherry-picking or refining involved in this piece. The dialogue and voiceover audio came directly from Veo 3, embedded with the clips themselves. The only exception was the score, which I generated using Udio. I trimmed and adjusted the music slightly to tighten the timing and help the cuts sync up better. All in all, it took me about two hours of playful prompting to generate and assemble everything. A brief but exhilarating ride.

---