The Shot We Didn't Take
The photo shoot used to be an event.
You'd block out a half-day minimum. Book the location or arrange access to the property. Coordinate with the photographer — confirm availability, negotiate day rate, discuss concept. Source props. Brief a stylist. Sometimes do a casting for lifestyle shots. Drive to the location. Set up. Shoot. Review on the back of the camera, reshoot what didn't work, wrap. Ingest the files. Wait for the edit. Review the edit. Request adjustments. Wait again.
For a real estate project, we're talking about a week from brief to final files. For a product campaign, sometimes longer.
We still do real shoots sometimes. But the math has changed completely.
What NanoBanana changed
The first time I saw what NanoBanana 2 could do with a single reference image, I understood immediately that we were not looking at a better editing tool. We were looking at a different category of capability.
Take one decent photo of an apartment — not even a great one, just usable — and generate twelve variations. Different lighting conditions. Different styling. Different times of day. An angle the photographer never stood at because the furniture was in the way. A lifestyle composition with people in it, without a casting.
The quality isn't perfect on the first generation. That's the thing nobody tells you. The clients who walk in expecting to prompt once and receive a finished campaign visual are going to be disappointed. AI image generation is an iterative process — you refine, you regenerate, you layer in corrections. The difference is that the iteration loop is measured in minutes, not days.
What used to take a full production day now takes two to three hours of focused generation work. We deliver more. We invoice less per unit. And we take on more projects, because the pipeline moves faster.
The angle that didn't exist
Here's a specific example that crystallized it for me.
A real estate developer had a show apartment. Beautiful project — but the living room had a structural beam in an awkward position that made it difficult to photograph cleanly. Every angle either cut off a key design element or included the beam in a way that drew the eye. The photographer did their best. The results were usable but not ideal.
We ran the best photo through NanoBanana and generated a version with the beam repositioned, then variations from an angle that physically couldn't exist given the room's layout. The developer used those generated visuals in their sales campaign. Not because we were trying to mislead anyone — the beam is real, the apartment is real — but because the visual communication of the space didn't need to be limited by one bad positioning decision made forty years ago by an architect.
Nobody asked about the beam. Everyone liked the apartment.
The video side
Video has followed the same trajectory, maybe faster.
A client campaign that would have required a full video production — shoot day, crew, direction, editing, motion graphics — can now be assembled from a mix of generated visuals, stock adapted with AI, and AI-generated footage for the segments that don't require real performance. Not for everything. Not for branded hero content where the craft matters. But for supporting visuals, social cuts, product demonstrations — the volume content that used to be expensive to produce consistently.
We're producing more video than ever. With smaller crews, shorter timelines, lower unit costs. The creative work is still creative work — it still requires someone with an eye, a sense of composition, a concept. But the production ceiling has lifted significantly.
What it still can't do
I want to be honest about the limits, because the hype around AI image generation has created expectations that don't match the reality.
It cannot replace the energy of a real shoot. When you have a genuine campaign that needs a specific human moment — a real expression, a real interaction between real people in a real space — no generator gives you that. You feel the difference. The client feels it. The audience feels it, even if they can't name what's off.
It also cannot replace the art direction. A bad prompt gives you a bad image, faster. Two people given the same brief and the same tool will generate very different outputs — the one who has an artistic eye, who understands composition, light, and visual hierarchy, will consistently produce better results. The tool amplifies existing taste. It doesn't create taste from nothing.
And the iteration time, while much shorter than traditional production, is not zero. Clients who come in expecting a finished visual from a one-sentence brief leave frustrated. The ones who understand it's a dialogue — who are willing to review generations, give specific feedback, and iterate — get extraordinary results.
What we charge for now
What we used to charge for was time. Photographer day rate, editing time, production coordination.
What we charge for now is judgment. Knowing which tool to use for which brief. Knowing when to generate and when to shoot. Knowing how to get a generator to produce what the client actually needs instead of what the prompt literally says. Knowing when the output is good enough and when it needs another pass.
The commodity is the generation. The value is the eye behind it.
That hasn't changed. If anything, it's become more true.
The shot we didn't take still requires someone who knows what they're looking for.