Three rules for better ChatGPT Image prompts
PeekPrompt editors
1. Describe the scene in sentences
GPT Image models follow clear sentences better than long comma-separated tag lists. Two or three sentences that answer who, where, doing what and in which light give steadier results. State the format first: aspect ratio, single frame or collage, and whether there should be text.
2. Say what must stay the same
With a reference photo, the model guesses what to keep. One sentence such as “keep the facial features, hairstyle and glasses exactly” or “do not change the bottle shape or label” removes the guesswork. Then describe what should change.
3. Fix things in the same chat
Rather than starting over, ask for one change at a time in the same conversation: a darker background, a corrected word. Small steps keep the parts that already work.
