Photography, with new tools

A tool, not an author

Every leap in photography was met with the same argument: this is the end of the craft. Roll film. Colour. Digital sensors. The computer. None of them ended photography. They changed who could work, how fast, and what was possible on a given day.

AI is the current leap. It cuts the dead time that used to vanish into cloning, masking, and salvage. It lets a studio try ideas that would have needed a stage, a city, or a budget that was not available. The picture still starts with a decision: where to stand, when to wait, whom to trust.

The sitting is still the sitting

A generated face can be technically perfect in seconds. It cannot know why a look in the room is worth keeping. It has no timing. It does not feel the air change when someone stops performing.

The photographic eye is not a filter. It is attention: light, distance, the half-second before someone drops the pose. That is still a human job. Software can finish edges. It cannot hold a conversation, or decide that this frame, not the sharper one, is the picture.

Where it sits in the work

Inventing people and calling that a portrait sitting is not photography. AI belongs where it is useful. The camera belongs where it matters.

• Before: Mood boards for light, wardrobe, and location. A shared picture of the day, before anyone is on set.
• After: Cleaner files, quieter backgrounds, extended frames, colour that holds. Faster than the old retouch loop, without sanding the skin into plastic.
• Beside: When a real portrait needs a world that does not exist on the street that afternoon: a hybrid, labelled as such.

If the assignment is a person, there is still a meeting. Light still has to be placed. There is still a moment for the shutter.

Retouching

High-end retouch used to mean a long night of pixel work. The new tools are faster and, used with taste, more precise. The brief is not to invent a different face. It is to bring the file to the level the sitting already had in the room.

Generative fill can lift a stand, a cable, a stranger at the edge. Skin stays skin. Grain stays grain. If the picture starts to look generated, the retouch has gone too far.

Generation

Image models such as Midjourney and Firefly are useful when the job is a world, not a witness. That work is illustration with photographic language, not a substitute for a sitting that took place.

• Concept: Lighting, styling, and framing, agreed before the first frame is shot.
• Hybrid: A real portrait placed in a setting that could not be booked. The person is real. The room may not be. That difference is the point of labelling.

What it will replace

Work built on generic pictures, not on a specific person in a specific hour, is already shifting.

• Stock: A happy office, coffee on a table, a handshake in golden hour. No one needs to hire a crew for that.
• Catalog: A bottle, a shoe, a packshot with perfect light. The studio still wins when the object has to exist and the brand has to own it. The rest gets calculated.
• Generic portraits: A clean LinkedIn headshot from three selfies. Fine for that use. It is not a portrait sitting.

Editorial, campaign, and portrait work that depends on a real encounter does not vanish. It becomes clearer. The assignment is the person, not the pixels.

Truth, labels, trust

Photography was always a selection: crop, light, moment. AI is a further selection. The line is simple.

• Document: A real meeting. AI may denoise, grade, or lift a distraction. The event still happened.
• Optimized: A photograph, finished with tools: sharpness, light, careful retouch. No new objects, no new places.
• Enhanced: A photograph with generative work in it: an extended wall, a removed car, a sky that was not that sky.
• Generated: The image comes from a prompt or a model, wholly or for the most part. Illustration, not evidence.

If a picture would change what you believe about a person or an event, it is labelled. That is not a trend. It is the minimum.

Likeness and consent

A model of a face is not a sitting. A likeness of a real person is not built without a clear yes: not for a mood board that leaves the studio, not for a campaign file, not for a hypothetical.

The same rule applies to clients. If the face belongs to a person who exists, that person is in the work, or the character is generated and is not that person. Mixing the two without saying so is the line that does not get crossed.

Rights

Commercial work runs only through tools that are cleared for that use: licensed accounts, documented terms. Usage rights on a finished file are agreed in the same way as on a traditional production.

Training data is not a grey area to ignore. If a tool cannot stand behind commercial use, it does not enter a client job.