Can AI Really Deliver at the Level I Expect?

Making something impressive with AI is easy. Making something I’d actually put in front of a client is a completely different test.

Once I realized AI could potentially solve some of the production problems I was running into, I had another question:

Could it actually produce work at the level I wanted?

Not “good for AI.”

Not something that makes you stop scrolling because you can’t believe a computer made it.

I mean genuinely good creative work—the kind where, if nobody told you AI was involved, you wouldn’t be looking for the giveaway.

So I started testing.

A lot.

I’ve experimented with image and video tools from OpenAI, Midjourney, Seedance, Runway, Kling and Adobe, among others. Some became part of my regular workflow. Some were incredible at one very specific thing and frustrating at another. Some produced results I never expected. Others looked amazing at first glance and completely fell apart the second I looked closer.

And that last part became really important.

The first impression isn't the test

AI is very good at making something that looks impressive for three seconds.

Beautiful lighting. Incredible locations. Perfectly styled people. Cinematic camera angles. Things that would normally require an entire production crew can suddenly appear on a screen.

That part is exciting.

But commercial creative has to survive the second look.

Does the product actually look like the product?

Is the label correct?

Does the person still look like the same person in the next image?

Do the hands make sense?

Are reflections behaving properly?

Does the fabric look real?

Does the jewelry actually connect?

If I move the camera, does the environment suddenly change?

Can I make a tiny adjustment without accidentally changing five other things?

Those became much more interesting questions to me than simply, “Can AI make a pretty picture?”

Because yes.

It absolutely can.

Control matters more to me than surprise

One of the things I learned pretty quickly is that a beautiful unexpected result isn't always useful.

Sometimes AI gives you something better than what you asked for.

That can be amazing when you're exploring ideas.

But when I’m executing a specific creative direction, I don't necessarily want to be surprised.

I want control.

If the jacket needs to be blue, it needs to stay that exact shade of blue.

If the bottle is sitting three inches to the left, I don't want the label redesigned when I move it.

If I've established a particular person, environment or visual language for a campaign, I need to be able to continue building from it.

That’s where my testing started shifting from “look what AI can create” to “how reliably can I direct it?”

Those are two very different things.

I also stopped looking for one perfect tool

For a while, I assumed eventually I would find the one.

The AI platform that could do everything.

I don't really think about it that way anymore.

Different tools have different strengths, and those strengths change surprisingly fast.

I might develop an idea in one place, build or refine an image somewhere else, use another tool to create motion, and then bring everything back into Adobe for compositing, retouching, color, cleanup and final production.

Sometimes the smartest solution isn't one spectacular generation.

It's five different tools doing five different jobs.

And oddly enough, that started feeling very familiar.

Traditional creative production works the same way.

You don't expect the photographer to also be the stylist, set builder, makeup artist, retoucher, animator and colorist simply because they all contribute to the same final image.

Why would I expect one AI model to do everything?

The final 10% is where things get interesting

This is probably the biggest thing I've learned.

Getting something almost right has become incredibly easy.

Getting it exactly right is where the work is.

I can generate something beautiful and then spend far more time fixing a bottle, rebuilding a hand, correcting a reflection, matching skin texture, adjusting typography, compositing a real product, refining lighting or making a series of images actually feel like they belong to the same campaign.

And I'm completely fine with that.

Because my goal was never to type one sentence into a machine and have my job disappear.

My goal was to find a better way to execute ideas.

There’s a huge difference.

So, can AI really deliver at the level I expect?

Yes—with an asterisk.

I don't think AI is at its best when you ask it to replace the entire creative process.

I think it's at its best when you know exactly where to use it.

It can create environments I couldn't justify building.

It can let me explore ideas before spending money producing them.

It can make tiny budgets feel much bigger.

It can turn still images into motion, help extend sets beyond what physically exists, accelerate retouching and give me ways to approach a project that simply weren't practical before.

But I still have to know when something is wrong.

I still have to decide what looks good.

I still have to understand the brand, the product, the audience and the idea.

And sometimes I still have to photograph the thing.

That realization became a pretty important part of scemax.

We're not trying to prove that AI can do everything.

I'm much more interested in figuring out what it can do incredibly well, what still needs a human hand, and what becomes possible when you combine the two.

Because the standard isn't whether AI made it.

The standard is whether I'd put my name on it.