Last year I faked my kid’s stuffed deer.
Put him on a trip.
Used Gemini ads as cover for the experiment, never showed my four-year-old a second of Buddy the deer’s fabricated adventures. It was meant to be harmless. Or at least contained.
Turns out the line between “playful” and “full-on slop” is blurry.
The tools for making realistic video are too good. Too easy.
Google is pushing hard into this with Omni.
It’s a family of models promising to turn text, photos, video — anything — into anything else. Currently, it just makes video.
Omni Flash is out now inside Flow.
Veo still works. But Omni claims to know the world better. Characters stay consistent. Logic holds up.
Claims are cheap.
I tested it.
Buddy the deer went back to work.
Packed his AI bags.
The results were a disaster mixed with genuine talent.
Some clips were tight. Actually followed my instructions better than Veo did five months ago.
But the jump scares? Real.
Watched Buddy switch mid-air orientation while skydiving. Gravity forgot who it was working for.
Asked for artistic freedom next.
Told Omni to make a montage: Buddy packing, boarding a cruise, cute vibe. Asked for a funny suitcase item that pays off later.
The AI chose honey.
Later in the clip Buddy grabs it like sunscreen. Squirts sticky yellow stuff onto his hoof. “Uh oh.”
Not a bad joke.
But the bottle mutated.
Jar. Water squirt bottle. Honey squeeze tube.
Back and forth.
The final frame looked like the model vomited the last five seconds onto a canvas.
Chaos.
Text edits?
Google gives itself credit here. It works better than before.
Better is the key word.
Veo 3 refused to change things. I had to start over every time. Omni tries. Usually fails.
Told it to fix Buddy’s facial expressions. Got weird, melted faces.
Asked to remove antlers he shouldn’t have. Baby deer don’t have antlers.
He obeyed in one shot. Added antlers to every other scene in the clip.
Stubborn.
There’s a meter running.
Credits burn fast. 15 to 40 per scene. Edits cost another chunk.
I have the Pro plan. 1,000 credits a month.
Burned through them in 20 clips. 145 left.
If you want exactly what you imagine? You’ll bleed money chasing the model around corners.
I tried something personal.
Skipped Buddy.
Fed a neutral selfie into the grinder.
Told it: Make me eat spaghetti. Sit on a plane. Bite a baguette at the Eiffel Tower.
My stomach dropped when the video loaded.
Yes there are tells. The fork clinks too clean. A background woman duplicates on the flight.
Uncanny.
Convincing anyway.
Showed the pasta clip to my husband.
He knows I test tech. Didn’t tell him it was AI.
He watched me eat. Said it looked real. Only clue was the unfamiliar bowl.
He lives with me. Looks at me every day.
Didn’t clock it.
Other clips?
Social media bait.
One Eiffel Tower video looked slightly cartoonish. Another was sharp.
Until my AI self turned her head. Hair suddenly in a ponytail I don’t wear.
I know it’s fake.
Strangers might not. That makes me uncomfortable.
Exhaustion sets in.
Been shocked by deepfakes for years. Photos first. Video next.
Veo 3 got me last time. Omni should too. But the shock has faded to a dull hum.
Making cinema is still hard. Google sells a dream.
Omni is better. Recognizable improvements.
But the barrier is low.
Phone in hand. Credit card ready. You can sit in your living room and look like you’re in Maui with zero effort.
Are we at the singularity?
Maybe not.
We are definitely deep in the uncanny though.
All images and videos in this story were AI generated by Google
