Almost every generative brief arrives asking for photoreal, and almost none of them arrive having considered anything else. It is a default rather than a decision, and it is the most expensive default available: photoreal has the highest render count, the tightest gates, the largest post burden and the smallest margin for a wrong hand.
Why photoreal is expensive
The audience has spent their whole life looking at photographs. They know, without being able to articulate any of it, how skin behaves in light, how fabric folds, how a hand holds a cup, and what a real lens does at the edge of frame. Every one of those is a way for a frame to fail, and each failure is legible to somebody with no training at all.
A stylised register removes most of those tests. Nobody has a lifetime of reference for how a hand should look in a screen-printed illustration, so the frame is judged against the style’s internal consistency rather than against reality. That is a far more forgiving standard, and it is why the acceptance rate on stylised work runs several times higher.
The four questions that settle it
| QUESTION | IF YES | IF NO |
|---|---|---|
| Does the audience need to believe this happened? | Photoreal, and budget for it | Stylised is available and probably better |
| Is there a product whose appearance is the claim? | Photoreal for the product, at minimum | The whole piece is open |
| Will this run beside real photography of the same thing? | Photoreal, or the mismatch becomes the message | A distinct register is an asset, not a compromise |
| Is the category visually crowded with the same look? | Photoreal is the crowd. Consider leaving it | Photoreal may be the differentiator |
The distinctiveness argument
There is a commercial argument for stylisation that has nothing to do with cost. Generative photoreal has a house style — a certain smoothness, a certain symmetry, a certain quality of light — and a growing share of the audience can now identify it. Being recognisably in that register attaches your brand to everything else in it, including the worst of it.
A specific stylised register is defensible in a way photoreal is not. It can be documented, applied consistently, and owned. Nobody owns photoreal.
How to specify a register so it survives
A style name is a request. A list of properties is an instruction. Specify five things and the register holds across forty assets and three people:
- Medium: what it would have been physically made with. Screen print, gouache, cel, clay, pencil.
- Surface: what it sits on and how that reads. Uncoated paper, board, film emulsion.
- Mark: how the marks were made and how visible they are. Visible brush, hard vector edge, halftone dot.
- Palette: three or four named colours and nothing else.
- Absence: what this register does not contain. This line does more work than the other four combined.
The absence line is the one people skip and the one that produces consistency. "No gradients, no soft shadows, no photographic texture, no lens blur" is a register somebody can hit repeatedly. "Bold and playful" is not.
What stylisation does not solve
- Structure. A stylised piece with no story is a stylised piece with no story.
- Consistency across a sequence. A register still has to be locked and pasted; it is just more forgiving when it slips.
- Legible type, which fails identically in every register.
- Credibility for a demonstrated claim. If the audience has to see the product working, illustration cannot carry it.
- Disclosure obligations, which attach to synthetic content regardless of how it looks.
Twelve registers, each with the order to write the prompt in, the words that move the model, and the specific way that style fails.
PROMPTING BY ANIMATION STYLE →