AI Art Prompt Guide: How to Write Effective Prompts for Midjourney and DALL-E

Introduction: Why Prompt Craft Separates Good From Great AI Art

Anyone can type "a cat in space" into an AI image generator and get something back. The gap between that result and the striking, intentional artwork you see in portfolios comes down to one skill: prompt craft.

AI image models like Midjourney and DALL-E respond strongly to the specifics of your language. A vague prompt hands creative control to the model's defaults, which tend toward a generic, glossy look. A precise prompt directs the model toward a defined subject, style, composition, lighting, and mood, which is how you produce images that look deliberate rather than accidental.

This guide teaches a repeatable structure for writing AI art prompts, with concrete examples you can copy, paste, and adapt. It covers the two leading platforms, Midjourney and DALL-E, and the slightly different way each interprets prompts.

Who this is for: designers, marketers, content creators, hobbyists, and anyone who has tried AI image generation and wants more consistent, professional results.

What you will learn: a prompt formula that works across platforms, how to control style and lighting, platform-specific parameters, and how to debug prompts that produce disappointing images.

Prerequisites and Setup

Choose Your Platform

• Midjourney: accessed through Discord (or its web app for paid users). Best for artistic, painterly, and highly stylized imagery. Subscription required.
• DALL-E: accessed through ChatGPT or the OpenAI platform. Best for natural language conversations and quick iteration. Included with ChatGPT Plus.

You do not need both. Pick one to learn first, and the skills transfer.

Understand What a Prompt Actually Does

When you submit a prompt, the model breaks it into tokens and matches them against patterns in its training data. Words that evoke strong, specific visual references (named art styles, camera equipment, lighting setups) produce more focused output. Generic words produce blended, averaged results.

This is why "a cozy reading nook" gives you a forgettable image, but "a cozy reading nook, warm window light, dust motes in the air, watercolor illustration, muted earth tones" gives you something with a clear point of view.

Step 1: Learn the Universal Prompt Formula

Most strong AI art prompts follow a flexible five-part structure. You will not use every part every time, but keeping all five in mind prevents the most common mistakes.

1. Subject: what the image is about. Be specific.
2. Description: details about the subject, including action, expression, and surroundings.
3. Style: the medium, art movement, or artist reference.
4. Composition and lighting: camera angle, framing, and light quality.
5. Mood and color: emotional tone and palette.

The Formula as a Template

[Subject], [descriptive details], [style or medium],
[composition and lighting], [mood and color palette]

Weak Versus Strong Example

Weak:

a coffee shop

Strong:

a small independent coffee shop on a rainy evening,
warm interior glowing through fogged windows, a barista
wiping the counter, steam rising from an espresso machine,
cinematic photograph, shallow depth of field, golden
hour light, moody and intimate, amber and deep brown palette

The strong prompt is longer, but every phrase pulls the model toward a specific decision. Length is not the goal; specificity is.

Step 2: Master Subject and Description

The subject is the anchor of your image. Vague subjects produce vague results.

Be Specific About What You See

Instead of "a woman," write "a woman in her sixties with silver hair pulled into a loose bun, reading glasses on a chain, wearing a faded denim apron." Specificity gives the model concrete visual anchors.

Describe Action and Expression

Static subjects look lifeless. Add what the subject is doing and feeling:

a young engineer soldering a circuit board, brow furrowed
in concentration, magnifying loupe clipped to her glasses,
sparks reflecting in her safety goggles

Set the Environment

Place the subject somewhere with detail:

in a cluttered basement workshop at 2 a.m., blue monitor
glow, half-empty coffee cups, schematics pinned to a corkboard

Step 3: Control Style With Medium and Art References

Style is where beginners lose control. Without a style cue, the model defaults to a generic photographic look. Naming a medium or art movement dramatically changes the output.

Photography Styles

cinematic photograph, 35mm film, grainy, kodak portra 400,
shot on Leica, photojournalism, golden hour

Illustration and Painting Styles

flat vector illustration, limited palette, geometric shapes,
mid-century modern poster style, bold outlines

oil painting, thick impasto brushstrokes, in the style of
post-impressionism, visible canvas texture

Specific Artistic References

You can reference named artists, art movements, or franchises, though results vary and the ethics of directly copying living artists is worth considering. Safer alternatives reference movements or aesthetics:

art nouveau poster aesthetic, Alphonse Mucha influence,
ornate floral borders, muted vintage palette

Medium Cues That Work Well

Useful style keywords to keep on hand:

• Photo: cinematic, editorial, documentary, macro, aerial, tilt-shift
• Illustration: flat design, line art, watercolor, ink wash, children's book illustration
• 3D and render: octane render, unreal engine, isometric, low-poly, claymation
• Painting: oil painting, gouache, impasto, fresco, sumi-e ink

Step 4: Direct Composition and Lighting

Composition and lighting are the difference between an image that looks designed and one that looks random.

Composition Keywords

• Framing: close-up portrait, wide establishing shot, extreme macro, overhead view
• Angle: low angle, bird's-eye view, dutch angle, eye-level
• Arrangement: rule of thirds, centered composition, symmetrical, leading lines
• Depth: shallow depth of field, deep focus, bokeh background

Lighting Keywords

Lighting defines mood more than any other element:

• Natural: golden hour, overcast, soft window light, dappled shade
• Studio: softbox lighting, three-point lighting, rim light, high-key
• Dramatic: chiaroscuro, Rembrandt lighting, single light source, silhouette
• Atmospheric: backlight, volumetric light rays, neon glow, candlelight

Example Combining Composition and Lighting

a single origami crane on a vast empty desk, extreme
close-up, shallow depth of field, soft morning window light
from the left, long shadows, dust particles visible in
the light beam, minimalist composition, muted palette

Step 5: Use Platform-Specific Features

Midjourney and DALL-E interpret prompts differently and offer different controls.

Midjourney Parameters

Midjourney appends parameters after double dashes. The most useful:

a serene mountain lake at dawn, cinematic photograph,
wide angle, soft mist --ar 16:9 --stylize 250 --v 6

Key parameters:

• --ar 16:9 sets aspect ratio (try 1:1, 3:2, 9:16 for portrait).
• --stylize 250 controls how strongly Midjourney applies its aesthetic (range roughly 0 to 1000; higher is more stylized).
• --chaos 25 increases variation between the four generated images (0 to 100).
• --no text uses negative prompting to exclude elements.

DALL-E Approach

DALL-E (through ChatGPT) does not use parameter flags. Instead, it responds to conversational refinement. Describe what you want, then ask for changes in plain language:

Generate an image: a serene mountain lake at dawn, cinematic
photograph, wide angle, soft mist, 16:9 aspect ratio.

Then iterate:

Make it more melancholic, cooler color palette, and remove
any boats from the scene.

DALL-E excels at this conversational back-and-forth, while Midjourney excels at precise stylistic control via parameters.

Step 6: Iterate and Refine Like a Pro

No prompt is perfect on the first try. Professional AI artists iterate.

Generate in Batches

Always generate four variations at once and look for the seed of a good image, then refine from there. Both platforms offer variation and upscale controls.

Change One Variable at a Time

When an image is close but not right, change one element per iteration. Change lighting, then composition, then palette. Changing everything at once makes it impossible to learn what works.

Keep a Prompt Journal

Save prompts that produced images you loved, with notes on what each phrase contributed. Over time you build a personal vocabulary of cues that reliably produce your preferred aesthetic.

Use a Reference Image When Available

Midjourney supports image prompts, where you prepend an image URL to influence the result. This is invaluable for matching an existing brand style or visual direction.

Tips and Best Practices

Front-load the most important words. Models weight early tokens more heavily. Put your subject first, decorative details later.

Avoid contradictory instructions. "Photorealistic cartoon" or "bright moody lighting" confuses the model. Pick a direction.

Use negative prompts deliberately. Telling the model what to exclude ("no text, no watermark, no people") often improves results more than piling on positive cues.

Steal color palettes from references. Name specific palettes ("terracotta and sage," "muted pastels," "high-contrast black and white") instead of vague "nice colors."

Match the medium to the message. A corporate logo wants flat vector style; an emotional editorial piece wants cinematic photography. Choosing the wrong medium undermines the concept.

Common Mistakes to Avoid

Prompt stuffing. Adding thirty keywords does not produce a better image. It produces a confused one. Edit ruthlessly to the cues that matter.

Ignoring aspect ratio. A landscape scene forced into a square looks cramped. Set the aspect ratio to match the subject.

Relying on "4k, hyperrealistic, trending on ArtStation." These were useful years ago but now mostly produce a generic glossy look. Describe the actual qualities you want instead.

Forgetting composition. Without composition cues, the model places the subject dead center every time. Specify framing and angle for intentional results.

Skipping iteration. The first generation is a draft. Treat it as such and refine.

FAQ

How long should an AI art prompt be?

Anywhere from one focused sentence to a short paragraph. Aim for a sentence per concept: subject, style, composition, lighting, mood. Twelve to thirty carefully chosen words usually outperform either a five-word prompt or a hundred-word prompt.

Which is better for beginners, Midjourney or DALL-E?

DALL-E is friendlier for absolute beginners because it accepts natural language and lets you refine conversationally. Midjourney rewards learning its parameter syntax but offers finer stylistic control. Try DALL-E first if you have never generated an image; switch to Midjourney when you want more artistic range.

Can I use AI-generated images commercially?

It depends on the platform and your subscription tier. Midjourney generally grants commercial rights to paid subscribers. DALL-E through ChatGPT generally allows commercial use, but you should verify the current terms on the platform you use and avoid directly imitating identifiable living artists or copyrighted characters.

Why do my images look generic even with a long prompt?

Usually because the prompt uses generic adjectives ("beautiful," "stunning," "amazing") instead of specific visual cues. Replace vague praise words with concrete descriptions of subject, style, lighting, and palette, and the images will immediately look more intentional.

Conclusion and Next Steps

Writing effective AI art prompts is a learnable skill, not a mystery talent. The formula of subject, description, style, composition, lighting, and mood gives you a reliable structure, and platform-specific parameters let you fine-tune the result.

Your next steps:

1. Pick one platform and learn its basics this week.
2. Take three of your past prompts and rewrite them using the formula in this guide.
3. Generate variations of each, changing one variable at a time, and keep the best.
4. Start a prompt journal documenting the cues that produce images you love.
5. Build a small personal library of style, lighting, and composition keywords to draw from.

Within a few weeks of deliberate practice, you will move from hoping the model gives you something usable to confidently directing it toward a specific vision. The models are powerful; prompt craft is how you actually wield them.