Home › Prompt Library › Prompt for DALL-E

Prompt for DALL-E

Write DALL-E prompts using natural, descriptive language that plays to its strengths.

Images · Works great in DALL-E

Copy-ready prompt

Write a DALL-E image prompt for [subject].
Describe in natural, complete sentences: the subject, setting, style, lighting, mood, and composition.
Rules:
- DALL-E responds well to descriptive natural language, not just keyword lists.
- Be specific about style and perspective.
- Avoid contradictory instructions.
Give 2 variations.

Want a version tailored to you?

Answer a few quick questions and the Image Prompt Generator builds a custom prompt from your exact details.

🎨 Open the Image Prompt Generator

Why DALL-E wants sentences, not keywords

DALL-E, and especially the version built into ChatGPT, is trained to understand language the way a person writes it — as full descriptive sentences with grammar and relationships between things. This is the opposite of Midjourney, which thrives on comma-separated keyword fragments and parameter flags. If you feed DALL-E a keyword dump like "cat, hat, sunset, cinematic, 4k," it will still make something, but it loses the connective information that tells it how the elements relate. Written as a sentence — "a fluffy orange cat wearing a tiny wool hat, sitting on a windowsill as the sun sets behind city rooftops" — DALL-E can place the cat, understand it is wearing the hat rather than standing next to one, and set the scene coherently. The prompt above is built around this strength: it asks for complete sentences covering subject, setting, style, lighting, mood, and composition.

Being specific without contradicting yourself

Specificity is what turns a generic image into the one in your head. Name the style explicitly — "in the style of a 1950s children\'s book illustration," "as a photorealistic product shot," "as a soft pastel watercolor" — because a bare description defaults to a bland, middle-of-the-road render. Describe the perspective too: "shot from a low angle looking up," "a flat top-down view," "an extreme close-up." The critical rule is to avoid contradictions. DALL-E cannot resolve "a minimalist scene with intricate ornate detail everywhere" or "a bright cheerful photo with a dark ominous mood," and when it tries, you get muddy, confused output. Pick one clear direction per attribute and commit to it.

Handling text, counts, and the things DALL-E struggles with

Certain requests are genuinely hard for the model, and knowing them saves frustration. Rendering readable text inside an image is unreliable — short words sometimes work if you put the exact string in quotation marks ("a sign that reads \'OPEN\'"), but long phrases usually come out garbled. Exact object counts beyond about three or four are inconsistent, so "a dozen apples" may give you eight or fifteen. Precise spatial relationships between many objects can drift. When these matter, describe them plainly and be prepared to regenerate, or simplify the scene so fewer constraints compete. Because DALL-E in ChatGPT is conversational, you can also just ask for fixes in plain language: "make the lighting warmer" or "move the subject to the left and add more sky."

Iterating through conversation

The single biggest advantage of DALL-E inside ChatGPT is that image generation is a dialogue, not a one-shot command. Start with your full descriptive sentence, look at the result, then refine incrementally: "same image but at golden hour," "keep the composition but switch to an illustrated style," "remove the background clutter." Because the model remembers the prior image in context, these edits build on what you already have instead of starting over. This makes DALL-E especially good for people who know roughly what they want but need a few rounds to dial it in, and it rewards clear, plain-English feedback more than clever prompt syntax.

Why this prompt works

DALL-E interprets natural descriptive sentences better than terse keyword lists. This prompt writes in full descriptive language while still specifying style, lighting, and composition — the way to get coherent results from it.

How to customize it

Use full descriptive sentences rather than keyword dumps.
Be explicit about style and perspective.
Avoid contradictory details that confuse the model.

Example output

Sample only

Prompt sent to DALL-E:

"A cozy independent bookstore on a rainy autumn evening, photographed as a warm, realistic interior shot. Soft golden light spills from a vintage lamp onto tall wooden shelves packed with books, while rain streaks the front window and blurs the streetlights outside. The mood is calm and inviting, and the composition looks down a central aisle toward a small reading nook with a worn leather armchair."

What it produces: A photorealistic, symmetrical shot receding down the aisle, with warm lamplight contrasting the cool blue rain outside the window. The armchair anchors the far end as a clear focal point, and the overall feel is snug and atmospheric rather than clinical.

Variation A (illustrated):

"The same cozy bookstore scene, but reimagined as a soft, hand-painted watercolor illustration with visible brush texture, gentle muted colors, and a storybook feel."

Variation B (different time and angle):

"The same bookstore, but on a bright sunny morning, shot from a low angle near the floor looking up at the shelves, with sunlight streaming through the window and dust motes in the air."

Prompt variations to try

Product mockup

Write a DALL-E prompt for a clean product photo of [product]. Describe it in natural sentences: the product placed on a specific surface, the background (solid color or subtle setting), studio lighting with soft shadows, and a straight-on or three-quarter angle. Ask for a photorealistic, commercial style. Avoid clutter and contradictory details. Give 2 variations with different backgrounds.

Character illustration

Write a DALL-E prompt describing a character: [character]. In full sentences, describe their appearance, clothing, expression, and pose, then the background and mood. Name a clear illustration style (e.g. flat cartoon, semi-realistic digital painting) and the framing (full body, waist-up). Keep the style consistent and non-contradictory. Give 2 variations with different poses or outfits.

Scene / landscape

Write a DALL-E prompt for a landscape of [place]. Describe in natural sentences the location, the time of day and weather, the lighting and its color, the mood, and the composition (foreground, midground, distant horizon). Specify whether it should look like a photograph or a painting. Give 2 variations set at different times of day.

Common mistakes to avoid

Feeding DALL-E a comma-separated keyword list. It is tuned for natural language; rewrite fragments as a full sentence so it understands how the elements relate.
Including contradictory attributes such as "minimalist yet highly detailed" or "cheerful but dark." Pick one clear direction per attribute or the model produces muddy results.
Expecting long, accurate text in the image. DALL-E renders text unreliably; keep it to short strings in quotes and regenerate, or add the text later in an editor.
Leaving the style unspecified. Without a named style, DALL-E defaults to a generic look. Say explicitly whether you want a photo, a watercolor, a 3D render, and so on.
Treating it as one-shot. In ChatGPT, DALL-E is conversational; instead of rewriting the whole prompt, ask for a small edit like "make it warmer" and it refines the existing image.

Frequently asked questions

Does DALL-E use parameters like Midjourney's --ar?

No. DALL-E does not use flag-style parameters. You control aspect ratio and other attributes by asking in plain language ("make it a wide landscape image" or "a tall portrait format") or, in some interfaces, by selecting a size option. Everything else is set through descriptive sentences.

Why does the text in my image look garbled?

Rendering readable text is a known weakness of diffusion image models. Short words placed in quotation marks sometimes work, but longer phrases usually come out misspelled or nonsensical. For reliable text, generate the image without it and add the words afterward in a design tool.

How do I get a specific number of objects?

Small counts up to three or four are fairly reliable if you state them clearly. Beyond that, DALL-E becomes inconsistent. If an exact count matters, keep the scene simple, name the number explicitly, and regenerate until it matches, or describe the arrangement more concretely.

Can I edit a DALL-E image after it is generated?

Yes. In ChatGPT you can ask for conversational edits ("change the background to a beach," "make the lighting cooler") and it will revise the existing image. Some versions also support inpainting, where you mark a region and describe only what should change there.

Tip: replace the parts in [square brackets] with your own details before you send. The more specific you are — audience, tone, goal, constraints — the better the AI output.