Prompts written in the language of camera departments produce better results than prompts written in ordinary description. The reason lies in how the training material was labelled.
Why the vocabulary works
Film and video have a precise established terminology for framing, movement and lens behaviour, developed because crews needed unambiguous instructions.
Training footage carries that terminology in captions, scripts, shot lists and the descriptions attached to stock libraries.
The model therefore associates specific terms with specific visual patterns, which makes those terms function as reliable controls rather than as vague suggestions.
Which terms carry the most weight
Shot size is the strongest single control. Wide, medium and close each produce a clearly different framing with little variance.
Camera movement is next, since a dolly, a pan, a handheld follow and a static locked frame are visually distinct and consistently labelled.
Lens terms work through their side effects. A long focal length compresses depth and softens the background, and the model reproduces those consequences whether or not it represents anything about optics.
Why adjectives underperform by comparison
Terms like cinematic, dramatic or beautiful describe an impression rather than a decision, and they appear in captions attached to wildly different footage.
The model averages across that range, which is why such words produce a generic result that is difficult to steer further.
Replacing one impressionistic word with two technical ones usually gives more control than adding three more impressions.
How lighting terms behave
Lighting vocabulary transfers well because it is also standardised: backlight, side light, practical sources, overcast diffusion and hard midday sun each name a specific look.
Naming the source and its direction gives more consistent output than naming a mood, since mood terms depend on colour, contrast and subject together.
Time of day works as a compact instruction for the same reason, carrying colour temperature, angle and softness in one phrase.
Where the vocabulary stops helping
Terms describing intent rather than execution do not transfer. A shot that builds unease is a judgement about a sequence, not a property of a frame.
Editing concepts fail similarly, because the model generates one clip and has no representation of what precedes or follows it.
The dividing line is whether the term describes something a camera operator could set up in isolation, and instructions on the wrong side of that line are better handled in the edit than in the prompt.