How we created our AI promo video

I created this 20 second promo video using Google's latest AI video generation tool: Veo 3.1. I generated the music with Google’s Lyria 3 Pro and did the narration myself. I then used the free version of Canva to add the text and logo, and finally stitched the audio and video together using iMovie. It took an afternoon, during which I was doing other things. Since it takes a few minutes to generate each group of videos, it helps if you have some side work to do simultaneously.

My tips for creating a video with Veo 3.1

1. It's frustrating. Maybe just knowing this going in will prepare you. You'll have videos that you can tell are unusable from the first instant. A woman with coffee cups in both hands, a person sitting across from their identical twin, somebody picking up their purse, only to have a purse remain in place. You need to be very specific in your prompt, and even then you need to keep your fingers crossed.

   At times, you can provide one direction explicitly, and Veo will do the exact opposite. For me, the reflection of the woman's face in the computer screen was a killer. It would be distorted, or too cartoonish, or have a mustache. And no matter what I tried, I couldn't get Veo to not show it. It's as if each time, it was saying, "Here's a video with a different reflection of the woman's face." Oh, the curse words did fly.

   Another error I ran into a lot was spelling. On the woman's screen, it says the words "Access Granted." I used that text after, "Successfully Signed In," was misspelled time and time again. If there's one thing you'd think AI would be really good at, it's spelling, particularly since I typed the words in the prompt. Unfortunately, not so.

2. It can get pricey pretty quickly. This video cost me $292 in total. Obviously, compared to producing an actual video, this is a bargain, but the amount caught me by surprise when I looked at my bill. An 8 second video costs $3.20 and you can generate 4 at a time. Of course, I only used two of the videos I generated, with an extensive cutting room floor. If I had been happy on the first two tries, it would have been super cheap. I think putting the whole thing together for under $50 would have been a reasonable goal had I not spent time pursuing aspects that really didn’t pan out.

   One thing to note is that if you can manage to create the audio yourself, turn off Generate Audio. It cuts the cost from 40 cents a second to 20, which means every time you click the Generate button the cost is $6.40 as opposed to $12.80. For a lot of my videos, I had Generate Audio turned on. And despite that, I ended up doing the audio outside of Veo.

3. Extending videos doesn't seem to work out great. Veo lets you generate 8 second videos. Then you can extend videos by 7 seconds at a time. My original idea for the video was to show the Access Granted screen, followed by zooming in on the phone for a second. Then I wanted to extend the video by having the camera pan out, showing the woman typing an email. The problem was that when the camera panned out, there were too many differences. Different woman, different hair, wearing sunglasses, in a turtleneck, with new people around her, on a similar - but not the same - street. I tried (too) many times to describe a cohesive scene, and no matter how specific I was, there were always glaring discrepancies.

However, despite these complaints, I am very pleased with the output. I think it looks professional, and it only took an afternoon. In 6 months the process will probably be easier to use, less expensive, and with less errors. For now, I'd say it's very impressive, and if you have a clear idea what you want to do, it's worth trying out - just be prepared that it can be aggravating.

Here are the prompts that I ended up using.

For the first 8 seconds

A realistic, relaxed, 30 year old woman wearing a lightweight, white top and jeans is sitting at an outdoor cafe in Paris. She has a cup of coffee and is looking at her laptop screen. It is the golden hour. The camera zooms in from a wide angle to over her shoulder to show the computer screen is on and has a web browser open that then pops up with the message, "Access Granted." She is looking at her screen but not typing. Next, pan to her leather purse which is hung over the back of the chair. The purse has just the top corner of her phone peeking out of it. Make the phone very subtle. The video is one continuous shot. Avoid all reflections.

For the extension

Continue to focus on the phone for three seconds, then fade to black.

For the music

Upbeat French piano jazz

If you have any questions or comments, please feel free to email me.