Key answer
You do not need to learn every technical term before asking. Everyday language, custom names and standard terms are all usable if they identify what you mean. Avoid leaving a word that could mean several different concepts undefined for this task.
Full explanation
Google recommends natural language rather than rewriting ordinary questions as academic papers. Official guide WinoDict tested something more specific: giving models unfamiliar words with definitions and asking them to solve tasks. Models used definitions, but performance depended on model and wording; a definition did not guarantee correct use. This supports explaining meanings, not assuming that every informal name will be recognized accurately. Study
For example, say “I want the person in the video to keep looking the same” without knowing “character consistency.” But “make the video more premium” remains ambiguous: specify lighting, color, composition or editing.
Accurate industry terms, device names, software features and research concepts usually make the target easier to locate. Do not choose a term you do not understand just to sound professional. Describe it first and ask AI for candidate names.
By “don't change the face,” I mean keeping one character's facial features consistent across shots, not replacing everyone's face with the same face. What is this requirement usually called? Distinguish similar concepts.
The definition constrains the task. Standardize the name afterward without letting a label decide the requirement for you.
References
1. Google, Tips to write prompts for Gemini, official user guide
2. Eisenschlos et al. (2023), WinoDict: Probing language models for in-context word acquisition, EACL