How DALL·E 3 reads a prompt
DALL-E 3 from OpenAI was designed to understand ordinary language, so you write prompts the way you would brief an illustrator. Describe the subject, what is happening, the setting, the style and the mood in complete sentences. It handles long, detailed prompts and multiple subjects well, and it follows instructions about placement and relationships more faithfully than many keyword-driven models.
There are no double-dash parameters. In ChatGPT you ask for a wide or tall image in words, and you refine results through follow-up messages such as "make the sky stormier". Through the API, size and quality are request settings. OpenAI has since introduced newer image models in ChatGPT, so the model behind image generation may differ from DALL-E 3; the same conversational approach still applies. Check OpenAI's documentation for current options.
The prompt formula
[medium/style] of [subject] [doing what] in [setting], [lighting], [composition], [mood], [format request]
Example
A warm watercolor illustration of a grandmother and her grandson planting tomato seedlings in a backyard garden, late afternoon sunlight casting long shadows, the two figures in the lower half with a wooden fence and sunflowers behind them, gentle and nostalgic mood. Make it a wide landscape image.
DALL·E 3 parameters and controls
| Control | Syntax | What it does |
|---|---|---|
| Aspect ratio (ask in words) | Plain-language request | In ChatGPT, ask for a wide, tall or square image. Through the API, set the size parameter instead.e.g. Make it a tall portrait image. |
| Follow-up edits | Conversational follow-up | Refine an image by describing the change you want in a new message instead of rewriting the whole prompt.e.g. Keep everything the same but change the season to winter. |
Do this
- Write complete sentences and describe the scene in a logical order: subject, action, setting, light, style.
- Name the medium explicitly ("oil painting", "35mm film photo", "flat vector illustration") to avoid a generic look.
- Describe the position of each subject when there are several ("on the left", "in the background").
- Request the format in words: wide, tall or square.
- Refine with short follow-ups that change one thing at a time.
- For text in images, quote a short phrase and describe where it appears.
Avoid this
- Pasting Midjourney parameters like --ar or --stylize, which DALL-E 3 ignores or treats as text.
- Listing disconnected keywords instead of describing a coherent scene.
- Requesting images in the style of living artists or copyrighted characters, which the system may refuse or rewrite.
- Rewriting the entire prompt for a small change instead of using a follow-up instruction.
DALL·E 3 prompt examples
Storybook scene
A cozy storybook illustration of a family of foxes having a picnic under a giant mushroom, fireflies glowing at dusk, soft colored-pencil texture, warm golden and plum tones, gentle and magical mood. Make it a wide image.
Infographic-style
A clean flat vector illustration showing the water cycle: a lake on the left, clouds forming above mountains in the center and rain falling on the right, with simple arrows connecting each stage, bright blue and green palette, white background.
Realistic photo
A realistic photo of a home office corner with a wooden desk, a laptop, a trailing pothos plant and a mug of tea, soft morning light from a window on the left, shot at eye level, calm and tidy mood.
Surreal concept
A surreal digital artwork of a whale swimming through a sky full of clouds above a small coastal town, people on rooftops looking up in wonder, pastel sunrise colors, dreamy cinematic atmosphere. Make it a tall portrait image.
DALL·E 3 prompt FAQ
Does DALL-E 3 have parameters like Midjourney?
No. DALL-E 3 has no double-dash flags. In ChatGPT you describe what you want, including the format, in plain language. Through the OpenAI API, settings such as size and quality are passed as request fields rather than written into the prompt. Anything that looks like a Midjourney flag is ignored or may confuse the result.
Can I use a negative prompt with DALL-E 3?
There is no negative prompt field. You can say what to leave out in a sentence, but results are more reliable when you describe the scene you want positively, for example "an empty beach at dawn" rather than "a beach with no people". If something unwanted appears, ask for its removal in a follow-up message.
Why does ChatGPT change my prompt?
When using DALL-E 3 inside ChatGPT, the assistant may rewrite your request into a more detailed prompt before generating. This often helps, but can drift from your intent. To limit it, be specific and ask it to use your wording as written. Newer image models in ChatGPT may handle this differently, so results can vary.
How long should a DALL-E 3 prompt be?
One to four sentences works well for most images. DALL-E 3 handles long prompts, so add detail where it matters: the subject, the action, the setting, the lighting and the style. Avoid filler and conflicting instructions. If a detail keeps getting lost, move it earlier in the prompt or ask for it in a follow-up.