Google introduced Gemini Omni at I/O on May 19, 2026, and I was curious about one thing in particular: how much could it do if I gave it a vague prompt? Omni Flash can create video and audio, so I gave it as little direction as possible.

I decided to have it create a promo video for a random coloring book. No cover, no title, nothing. I expected something rough, maybe good enough to build on. Instead, Gemini gave me a 10-second promo that looked like the kind of video someone selling a coloring book could actually use. What it chose to do with that almost empty prompt was the part I didn’t see coming.

Gemini added a family I never asked for

A vague prompt gave it room to come up with a better idea than mine

I only gave Gemini Omni a vague prompt that told it to create a promo video for a coloring book. I never gave it details such as how many people to include or what scene to use. I didn’t even upload a cover or give it a name for the book. I wanted to see how creative Gemini Omni could get.

It understood the assignment on the first try and created a video with a little girl sitting at her desk with her coloring book. Then, it turned into a family thing, showing a dad with his son lying on the living room floor having fun coloring. All those details came from Gemini, since I never told it to add those people or have anyone lie down on the floor. That’s the way Google designed Gemini Omni, since its docs say it tries to build a multi-shot story from your prompt by default. I’m actually glad I didn’t give it more details because I don’t think I would’ve come up with a better scene.

I’m not saying that vague prompts are going to give me perfect results every time. However, it’s a great way to start when you’re not sure what you want. At least it gives you a starting point, and from there, you decide what to change.

I never told Gemini how to pace any of this

It still knew when to hold a scene and when to move on

I have to admit that Gemini spent just the right amount of time on each scene. Most scenes lasted around two seconds, while the close-up of the coloring page was closer to 1.5. The video didn’t feel unbalanced, and it’s not like it spent five seconds on the intro and rushed the rest of the video. I replayed it a few times, waiting for a scene to feel rushed or drag on, and none of them did. The shorter cut made sense since a detail shot doesn’t need as much time to register as a scene with people.

The narration was also spot on since it was short, to the point, and promotional without sounding like it was trying too hard. It didn’t have any distracting graphics that would’ve made the video feel too loaded. And while the video was only 720p, it looked clean enough that I didn’t feel like I had to apologize for the quality.

What makes this more useful is that I never told Gemini how to divide those 10 seconds. I didn’t say how long the opening should last or when it should move to the close-up. It handled that on its own. If the timing is off, you can tell Gemini when to switch scenes using regular language. If the video feels too busy, you can ask for a single continuous shot or say “no scene cuts.” Even with five scenes packed into 10 seconds, it didn’t feel like a bunch of AI clips stitched together.

The final frame was doing most of the selling

That made it the obvious place to test a simple title change

Gemini made another choice I probably wouldn’t have thought of: it saved the most polished part of the video for the final frame. I could see the coloring book from the beginning, but the first few scenes were about people using it, not showing off the finished result. Then the final shot showed a fully colored fantasy page with the title “Color Your World.”

The final sequence stayed up for around 2.5 seconds, longer than any of the other shots. That gave me enough time to notice the finished artwork and read the text Gemini had added. I can’t say it was following some kind of marketing strategy, but that’s how it seemed.

The video showed the experience first, then ended with the finished artwork and title, so that was the last thing viewers saw. If Gemini had placed the title at the beginning, the scenes that followed probably would’ve taken some of the attention away from it.

“Color Your World” wasn’t a title I had given Gemini either. It came up with that on its own. I liked it, but I wanted to see whether changing it would affect anything else in the video. I told Gemini to change it to “TEST TITLE” and keep the rest the same.

The new version stayed fairly close to the original, although it wasn’t identical. It’s good to know if you like most of the result but want to polish one part. Tell Gemini exactly what to replace, put the new text in quotation marks, and ask it to leave everything else alone. Then watch the full video again, not just the title, because one small edit may quietly affect something else. Google recommends keeping edit prompts simple and adding “Keep everything else the same” when changing one specific part.

Ten seconds was enough to make the idea worth keeping

I gave Gemini Omni one vague sentence and expected a rough clip I’d have to rebuild. Instead, it came back with an idea I know I wouldn’t have thought of; pacing that didn’t feel rushed, and an ending I only wanted to tweak. That surprised me. I gave it almost nothing and still got more than I expected.

Ten seconds won’t work for every kind of promo. Google says longer videos are coming, but even now, it’s enough to test an idea before spending time planning every shot. If you’re stuck and don’t know what kind of video to make, try one vague prompt first. You may get a direction you can build on.