I Tested Google's Veo 3 — Here's What Wowed Me and What Frustrated Me

(Image credit: Google Ultra)

Veo 3 was one of the headline moments at Google I/O 2025, billed as the company's biggest leap yet in AI video generation.

It arrives as part of the new Google AI Ultra subscription — the tier that unlocks Google's most capable models and premium tools. That bundle includes both Veo 3 and Flow, a companion tool that stitches clips together into something closer to a real cinematic sequence.

The pitch is straightforward: feed it a plain-text prompt and get back a hyper-realistic clip, complete with audio, dialogue, and sound effects. For casual creators, it's about as advanced as anything out there right now. After putting it through its paces, I can confirm the results are genuinely jaw-dropping at times — though the experience isn't without rough patches.

Here's what won me over, and what left me grinding my teeth.

What Veo 3 gets right

I've spent time with plenty of AI video tools, including earlier Veo releases, and this version is by far the easiest to work with when it comes to layering in sound and dialogue.

The realism is the standout. The fact that an 8-second clip can render in under two minutes on an ordinary computer — no crew, no studio — is remarkable on its own. People all over the internet are posting Veo 3 clips that are nearly impossible to tell apart from footage shot by humans. It's a vivid reminder of how quickly this technology is advancing.

Google Veo 3 realism just broke the Internet yesterday. This is 100% AI. 10 wild examples: 1. Street interview that never happened.

Where Veo 3 still struggles

For all that it does well, Veo 3 isn't seamless yet. It's still flagged as experimental, so Google is clearly ironing out the wrinkles — but these were the biggest snags I hit during testing.

1. Prompt interpretation is hit-or-miss

Veo 3 sometimes fumbles spatial instructions. When I asked for an overhead camera angle, I got a slightly tilted side view instead. It seems to favor cinematic style over literal accuracy, which chips away at how much creative control you actually have.

2. Audio doesn't always fire — and the reason isn't obvious

By default, Veo runs in Veo 2 mode, which has no audio at all. I only figured that out after a string of silent clips and some digging. To turn on audio and dialogue, you have to manually switch to "Experiential Mode" under the "Quality" tab.

Even once it's on, lip-syncing is inconsistent and dialogue sometimes vanishes entirely, like a poorly dubbed foreign film. The subtitles, meanwhile, are almost always wrong or misspelled.

3. Busy scenes trip it up

Veo 3 is at its best with a single subject. Push it toward longer or more elaborate scenes and things start to unravel — the storyline gets muddy and character interactions tend to feel stiff or repetitive. If you're hoping to build a multi-character, multi-scene narrative, keep your expectations in check.

4. The interface needs polish

(Image credit: Future)

At points the interface felt clunky and unstable. I ran into an unexpected session timeout that wiped out a generated video, with no recovery option in sight. And when I prompted the model to add dialogue to a scene, what came back didn't match the scenario at all. For a tool this powerful, the UX still feels unfinished.

5. It raises serious ethical questions

Veo's realism is astonishing — and a little unnerving. There's mounting worry that ultra-realistic AI video could erode the line between what's real and what's fabricated, especially as the tech gets easier to access. It also reopens thorny debates about authorship and originality in creative work.

Promising, but pricey

Veo 3 is a big stride forward for AI video, particularly for casual users who want fast, high-quality output. But at $249 per month — with a discount for the first three months — the Google AI Ultra package is a steep ask for a tool that still ships with notable bugs.

If you just want to experiment or knock out basic promotional clips, Veo 3 is exciting but not essential. For professional creators, it's worth keeping a close eye on. Just write tight prompts, keep your expectations grounded, and stay ready to hit re-render.