A prompt is the text the model works from. On tool pages it is written for you from the fields you fill in; open Prompt sent to the model to read it. In the studio you write it yourself. These habits save credits because you get closer on the first try.

Images: describe the shot

Write it the way you would brief a photographer or illustrator, in this order:

  1. Subject: who or what, with the details that matter. "A golden retriever puppy with a red collar", not "a dog".
  2. Action or pose: "sitting on a porch step, looking up".
  3. Setting: "wooden porch, autumn leaves, small town street behind".
  4. Light and time: "late afternoon sun, soft shadows".
  5. Style or medium: "35mm film photo", "flat vector illustration", "watercolor".
  6. Framing: "close-up", "full body", "wide shot from low angle".

Keep it to one scene. If you list two subjects doing different things, many models blend them.

For text inside an image, put the exact words in quotes: a poster that says "OPEN LATE". Some models handle lettering far better than others; see Choosing a model.

Editing a photo: say what changes

When you upload a photo, describe the change and what to keep: "Replace the background with a beach at sunset. Keep the person, clothes and pose the same." Short, specific instructions beat a full scene description.

Video: describe motion

A video prompt needs what moves and how the camera moves:

  • Motion: "she turns toward the camera and smiles", "waves roll in and break on the rocks".
  • Camera: "slow push in", "handheld tracking shot", "static tripod shot".
  • Sound, on models that make audio: "rain on the window, distant thunder".

Pick a length with the Duration setting. Video is priced per second, so try a short clip first and make a longer one once the prompt works.

Speech and music

  • Text to speech reads your text exactly. Write numbers and abbreviations the way they should sound, and use punctuation for pauses. The price is per 1,000 characters, so trim what you do not need.
  • Music: name the genre, tempo, instruments and mood: "lo-fi hip hop beat, warm vinyl crackle, 85 bpm".

What to leave out

  • Settings that have controls: aspect ratio, length and the number of images.
  • Long lists of quality words ("masterpiece, 8k, ultra detailed"). They add little on current models.
  • Names of real people you do not have permission to depict, and anything the content rules do not allow.

Iterate one change at a time

If a result is close, change one thing and generate again. Changing everything at once makes it hard to tell what helped.