So here's the punchline first: the campaign performed better than anything I'd manually produced in the past two months.
Now let me back up and explain why that's deeply unsettling.
The Part Where I Ruin My Own Lunch Break
It was a Tuesday. I had exactly 40 minutes before my next call, a lukewarm oat latte I'd already paid too much for, and a corner seat at the coffee shop I've been using as a second office since my apartment started feeling like a storage unit with Wi-Fi.
I wasn't supposed to be working on the ad project. I was supposed to be eating something. Instead, I was poking around with an AI Product Showcase Video Generator because a client had pinged me that morning asking for "a few variations" of a product showcase video for their skincare line — by Thursday.
"A few variations" in client language means: six formats, three aspect ratios, two tones (one aspirational, one relatable), and probably a last-minute request for a holiday version even though it's June.
I'd been putting off learning the A/B variant generation feature of these tools because, honestly, I assumed it would be like most AI "batch" features — technically functional, practically useless, producing twelve versions of the same beige mediocrity.
I was wrong. (I hate typing that.)
What I Was Actually Trying to Do
The original plan was simple: generate one clean product showcase video, export it, then manually duplicate and tweak it for different placements. That's the workflow I know. That's the workflow that has kept me employed.
But I fat-fingered a setting. Instead of selecting "single output," I somehow toggled into a batch mode I hadn't noticed before — I think I clicked the wrong dropdown while reaching for my coffee — and hit generate.
The tool started producing variants automatically. Different headline positions. Different pacing on the product reveal. One version opened with the texture close-up; another led with a lifestyle scene. The CTA button showed up at the 4-second mark in some, the 7-second mark in others.
I sat there watching the queue fill up like I'd accidentally ordered twelve things at a restaurant and the kitchen was already cooking.
The Part That Actually Impressed Me (Reluctantly)
Here's what I didn't expect: the variants weren't just resizes. The tool was making actual compositional decisions based on the aspect ratio.
The 9:16 version (vertical, for Reels/TikTok) had the product centered with text overlaid at the bottom third — correct for thumb-stopping scroll behavior. The 1:1 version pulled back slightly to give breathing room on both sides. The 16:9 version restructured the whole pacing, giving more time to the mid-shot because horizontal viewers apparently have longer attention spans, or at least that's what the layout implied.
I don't know if it was making those decisions intelligently or if I just got lucky with the defaults. Probably somewhere in between. But the output was — and I'm going to say this carefully — usable. Like, actually usable. Not "usable after two hours of cleanup" usable. Usable usable.
I sent three of the twelve variants to the client as a "preliminary look" (translation: I panicked and needed to buy time to figure out what had just happened).
They picked two of them. Said they looked "polished and on-brand."
I had made them by accident in a coffee shop while reaching for a drink.
The Uncomfortable Math
Let me be honest about what this means for my workflow — and maybe yours.
Before this, my process for A/B variant generation was:
- Create master version (2–3 hours)
- Duplicate and manually adjust for each format (45 min per variant)
- Export, review, fix inconsistencies (another hour)
- Total for 6 variants: roughly a full workday
What happened on Tuesday:
- Accidental batch generation: ~8 minutes of compute time
- Review and light editing: 35 minutes
- Total for 12 variants: under an hour
I'm not going to pretend that's not significant. It is. It's the kind of significant that makes you stare at your oat latte and think about what you're actually selling as a creative professional.
The tool I was using — I'd been exploring UGCVideo.ai for the product showcase video side of things — had this batch variant feature tucked away in a way that felt almost apologetic, like it knew it was going to cause an existential crisis and was trying to be polite about it.
What the Tool Doesn't Do (And Why That Still Matters)
Okay, let me be fair to myself for a second.
The variants were good. They were not great. There were things the AI couldn't do:
- It couldn't understand that this particular client hates the color yellow (long story involving a previous agency and a very bad campaign)
- It didn't know that the brand voice is "approachable expert," not "luxury aspirational" — two of the twelve variants leaned too polished, too cold
- The AI Character Maker used in the lifestyle shots had a consistency issue across variants — same general look, slightly different facial structure in two of them, which would have been a problem if the client had noticed (they didn't, but still)
These aren't small things. Brand context, client history, the subtle difference between "warm professional" and "glossy corporate" — that's the layer of work that still requires a human who's been in the room, read the brief three times, and sat through the feedback calls.
The AI is very good at generating plausible. It's not yet good at generating specifically right.
What I Actually Changed After This
I restructured my intake process. Now when a client asks for "a few variations," I ask them upfront: which platforms, which audience segments, what's the primary CTA. Not because I need the information to brief a team — but because I need it to brief the AI correctly before I let it run.
The quality of the output is almost entirely determined by the quality of the constraints I give it. Garbage in, twelve garbage variants out. Specific in, twelve actually-useful starting points out.
That shift — from "I make the thing" to "I define the parameters that shape the thing" — is the real skill change. And honestly? It took me embarrassingly long to accept that this is still a creative skill. It's just a different one.
So Here's the Question I Keep Coming Back To
The campaign went live. The two variants the client chose performed 34% better on click-through than the previous month's manually produced ads (their numbers, not mine — I'm not taking full credit for something I half-accidentally generated).
And I keep thinking: if the output is better, the turnaround is faster, and the client is happier — what exactly am I defending when I resist these tools?
Is it craft? Is it professional identity? Is it just the sunk cost of having spent years learning a process that's now being compressed into a dropdown menu I accidentally clicked?
Maybe the question was never "will AI replace creative work."
Maybe it's always been: what part of the work were you actually doing that mattered?
Top comments (0)