What it actually describes
The word comes from Japanese and refers to quality, not quantity. Two lenses set to the same aperture, framing the same subject at the same distance, will blur the background by the same measurable amount and still look nothing alike. One dissolves the background into cream. The other chops it into overlapping bright edges. That difference is bokeh.
It comes down to how a lens renders a point of light that is not in focus. In a well-corrected lens that point spreads into an evenly filled disc, and thousands of those discs overlap into a smooth wash. In a lens with strong spherical aberration the disc is brighter at its rim than at its center, so overlapping discs leave visible outlines, and a background of bare branches turns into a tangle of doubled lines.
Two other characteristic behaviours are worth recognising, because they are stylistic signatures rather than defects:
- Cat's eye. Toward the frame corners the highlight gets clipped into a lemon shape by the lens barrel. Strongest wide open, and part of why fast primes feel like fast primes.
- Oval highlights. Anamorphic optics squeeze the highlight into a vertical oval. This is the single most recognisable thing about anamorphic footage, more than the flares.
Where it earns its keep
Portraits are the obvious case: separating a face from a busy street is mostly a matter of destroying the street. Product work uses it to remove context. Night exteriors use it to turn ordinary streetlights into structure, since defocused point lights are the only way to make an empty frame feel full.
Backgrounds with small bright points are what create visible bokeh balls: fairy lights, wet pavement reflections, sunlight through leaves, traffic at dusk. A flat grey wall behind the subject has nothing to defocus, so the blur will be technically present and visually invisible.
Prompting it
Models know the word, and that is part of the problem: asking for it alone yields a generic soft wash. Specify the ingredients instead.
- ✅
85mm at f/1.4, string lights far behind the subject dissolving into large round bokeh - ✅
anamorphic 40mm, oval highlights in the background - ⚠️
blurry backgroundoften blurs the subject too, or produces motion blur - ⚠️
bokehon its own tends to give a uniformly soft backdrop with no highlight shapes in it
Naming the light source is the highest-value part of the prompt. The model cannot invent highlight shapes if the scene you described has no highlights.
Failure modes in generated images
Depth is inferred, not measured, so the blur boundary is where things break. Watch the edges of hair, glasses, and any hand held away from the body: these commonly end up on the wrong side of the boundary, sharp against a soft background or soft against a sharp one. Also check for a thin sharp outline hugging the subject, which is the same artifact that portrait mode on phones produces and the fastest way to spot a fake shallow look.
The other reliable tell is text. A sign 20 metres behind a subject should be an illegible smear. If it is readable, the blur was painted on rather than derived from the scene, and no amount of extra prompt language will fix that particular shot. Regenerate, or blur it properly in post.