Write Prompts That Actually Get What You Want

Lead with the concrete details a model cannot guess: subject, medium, mood, camera angle, colour palette, and length. "A weathered fisherman at dawn, backlit, 35mm film grain, muted teal and amber" gives the model far more to work with than "a nice fisherman photo." Nouns and adjectives do the heavy lifting; drop vague words like "beautiful" or "cool."

State the format and constraints up front so you spend fewer rounds correcting. Ask for a specific aspect ratio, a word count, a JSON shape, or "no text in the image." Telling the model what to avoid is as useful as telling it what to include, and negatives prevent the same mistake from repeating across generations.

Iterate instead of rewriting from scratch. Change one variable at a time so you can see what each edit does, and keep a short note of the phrasings that worked. On CinderHub you can run the same prompt across chat, image, and video models, so a prompt you refined for one medium becomes a reusable starting point for the next.