Talking AI babies is the latest generative trend – now I can’t stop making newborns talk
New parents curious about what they should expect from their newborn should probably steer clear of TikTok and AI. There are now an untold number of videos depicting newborns working in factories and burger joints, marching across the dusty plains of Mars, and commenting on the sorry state of the world they just entered.
Ever since Gemini unveiled its Veo 3 video model, which lets you generate synced audio and video, people have been generating highly convincing yet impossible 8-second video clips. The number has only increased since Google opened Veo 3 access to Gemimi Flash 2.5 Pro users.
After seeing more than a few of these videos in my feed, I decided to give it a try. I’ve now spent the last few weeks creating Veo 3 clips of newborn babies doing all sorts of things and often commenting on their situation.
Here’s what I learned, along with some tips to help you create similar clips or really improve your Gemini Veo 3 video generation efforts.
Planning
Unless you’re paying for an account, you get a maximum of three 8-second Veo 3 clips per day, so you want to use that number wisely and, if possible, pre-plan all three clips.
Start with a concept or idea
What do you want your newborn to be doing and saying? Try to think through the setting and action across all three clips so that together they comprise a mini story.
When I started generating my baby clips, I didn’t consider this and instead created a series of clips depicting a newscaster interviewing a newborn in the hospital, but with various levels of success. Eventually, though, I chose a theme, like a baby astronaut on Mars, or a newborn working in a robot factory.
Once I had a mini story in mind, my results improved exponentially.
Imagine a style
One of my favorite sets of clips is the one in which I imagined a baby in the role of Dr. Frankenstein. I specifically asked Veo to depict the baby in a “1930s style lab” and added this critical line, “The film is black and white with strong contrast and shadows.”
Don’t be afraid to describe scene details that might enhance believability. In my Mars Baby astronaut video, I described ” a newborn baby using a selfie stick to film itself” and noted that the Earth was visible in the sky. Veo 3 did an excellent job of including these details.
Be specific
I found that unless I described the main character as a newborn baby, I ended up with a toddler, and sometimes one that looked like a Pixar character.
Also, be sure to describe the action. What is the baby doing? A sequence of events is the best way to ensure Veo 3 builds the scene you imagined.
Keep the dialogue tight
You have just 8 seconds of video, so be sure to time your dialogue to make sure it’s not 15 seconds of blather. Whatever Veo 3 can’t fit, it will cut off.
Keep the characters to a minimum
One of the reasons these videos are so compelling is that they feature newborn babies speaking like adults. Achieving these results has occasionally proved challenging.
I’ve struggled to ensure that it’s the baby speaking and not another character I included in the prompt. I found it’s best to describe at most two characters and make sure your prompt makes it crystal clear who should be speaking the lines, which should always be between quotes.
Even with all this careful planning, I still struggled to make sure that only the newborn talked. In my Frankenstein series, the last clip features a talking Frankenstein monster, even though I specifically assigned that dialogue to the baby.
Tell a story
If you create all three clips in the same session, one after the other, Gemini Veo 3 will remember the style and even look of the included characters, which means you don’t have to repeat all those details in subsequent prompts.
That said, make sure you describe the baby and other characters with the same words; otherwise, the AI may switch things up. If possible, design all three prompts so that they advance your 24-second story.
Call it “realistic”
To ensure photo-realism, use the term “photo-realistic” or “realistic” in your prompt. Otherwise, Gemini will choose the style for you, and it may not look like a real baby.
Download the video
Gemini doesn’t save your generated videos indefinitely. They’re usually gone within a day, though the prompts will remain. If you like what the prompt created, be sure to download the video.
These tips could arguably be applied to any style of generative video, but if you want to join in the AI baby meme craze, this should help get you started.
