Generation controls

    Prompt

    A prompt is the written instruction a generative model reads to decide what to make — the one input almost every model requires.

    Models do not read a prompt as a checklist. The text is converted into a numeric representation and the whole of it steers the generation at once, which explains the two behaviours that surprise people most: word order carries real weight, and long prompts dilute rather than accumulate. Terms near the front tend to dominate; the fortieth adjective competes with the first thirty-nine.

    What a prompt should contain depends entirely on what else is going in. With nothing but text, the prompt owns subject, framing, light and motion. With a starting image, the picture already answers three of those, and a prompt that repeats them fights the image instead of directing it.

    There is a hard ceiling too. Models truncate at a maximum length rather than error, so an over-long prompt does not fail — it silently loses its tail, which is where people tend to put the thing they cared about most.

    In practice

    • Front-load the subject and the action; put style and grade at the end where truncation is survivable.
    • One clear noun beats three competing ones — a model asked for two subjects will often merge them.
    • Negations rarely work in a positive prompt; that is what the negative prompt field is for.

    The mistake to avoid

    Porting a prompt between models unchanged. Each model was trained on different caption vocabulary, so a phrase that is a style keyword to one is meaningless noise to another.

    Go deeper

    AI Image Prompt Engineering in 2026: The Complete Guide to Better Outputs

    The prompt patterns that actually work in 2026 for Flux, Midjourney, Ideogram and Imagen — structure, modifiers, negative prompts and the mistakes that produce generic output.

    Where you will run into it

    Related terms

    The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.