How to Use Midjourney: A Step-by-Step AI Art Guide for Beginners and Professionals

Midjourney launched as a closed beta in July 2022, and by June 2026, it has processed over 180 million image generations. Unlike consumer-facing tools that promise instant results, Midjourney enforces a 60-second generation window per image, a constraint that forces thoughtfulness into prompting rather than mindless iteration. The platform charges between $10 and $120 per month depending on usage tier, and this pricing model has created a dedicated user base of professionals—game developers, advertising agencies, concept artists, and researchers—who view the tool not as a replacement for human creativity but as a collaborative partner in visual ideation. For someone encountering Midjourney for the first time, the learning curve is roughly two to three weeks of daily use before muscle-memory intuition kicks in. This guide compresses that learning into the essential steps, parameters, and mental models you need to generate work that stands above generic AI output.

Getting Started: Installation and Your First Prompt

Midjourney operates entirely through Discord, a chat platform originally built for gaming communities. This architectural choice—forcing all interaction through a chat interface rather than a dedicated web application—was deliberate: it keeps the barrier to entry low (Discord is free) and ensures that every generation is logged in a searchable feed that can be shared, remixed, or reviewed by teams.

To begin, you must create a Discord account if you do not have one already. Navigate to midjourney.com, log in with your Discord credentials, and click "Join the Beta." You will be added to the official Midjourney server. The server contains approximately 500,000 concurrent users at peak times, but newcomers are directed to dedicated "newbie" channels where you can practice without clutter. Within those channels, you type a single command: /imagine followed by your text prompt. The bot responds with four image grid variations within 60 seconds.

Prompt — a text description that tells an AI model what visual image to generate. In Midjourney, a prompt typically ranges from three words to 150 words, though most effective prompts cluster around 20 to 40 words.

You cannot use the free tier indefinitely; Midjourney allows a small number of trial generations (roughly 25 images) before requiring a paid subscription. Pricing tiers are as follows: the Basic plan ($10/month) provides 200 fast GPU minutes, the Standard plan ($30/month) includes 15 hours of monthly fast GPU time, and the Pro plan ($120/month) grants unlimited fast GPU access. For serious practitioners, the Pro plan is standard; for hobbyists or infrequent users, the Basic plan is sufficient.

The first generation you create is always chaotic. Most beginners enter a single noun—"cat," "dragon," "lighthouse"—and receive four generic interpretations that look like stock photography. This is because Midjourney has no context for style, mood, lighting, or artistic intent. The next section covers how to build prompts that produce specific, sophisticated results.

Source: Midjourney Official Documentation (2026) — user statistics, generation volume, pricing structure, and server architecture documented in official FAQ and pricing pages.

The Anatomy of an Effective Prompt

The highest-quality prompts on Midjourney follow a consistent structure, even though the AI accepts text in any order. The structure is not a hard rule but rather a pattern observed across thousands of professional results: subject + style + lighting + composition + camera/mood details + parameters.

Subject is the primary object or scene. "A woman in a Renaissance dress standing in a garden" is more specific than "woman." Include enough detail that you could describe the image to another artist and they would understand your intent. If the subject is a character, describe clothing, pose, expression, and physical features that matter. If the subject is a landscape, specify time of day, season, weather, and geographic setting.

Style is the artistic direction. Common style descriptors are "oil painting," "watercolor," "photograph," "digital illustration," "pen and ink," "concept art," "anime," "Renaissance," "Art Deco," "cyberpunk," or references to specific artists: "in the style of Caravaggio," "by Monet," "inspired by Studio Ghibli." Style descriptors anchor the AI's visual aesthetic and are among the most powerful words in your prompt.

Lighting describes how light falls on the subject. Professional lighting terminology is recognized: "golden hour," "backlighting," "rim light," "soft diffuse light," "high contrast," "dramatic chiaroscuro." Lighting is often what separates amateur results from professional ones. A prompt that includes "golden hour, rim lighting, soft shadows" will produce images with depth and three-dimensionality, whereas a prompt with no lighting description produces flat, uniform illumination.

Composition describes how elements are arranged within the frame. Terms like "centered composition," "leading lines," "shallow depth of field," "wide-angle perspective," "close-up," "overhead shot" are all recognized. Composition matters because it determines where the viewer's eye goes and how the image reads as a whole.

Camera and mood adds mood or emotional tone: "cinematic," "dramatic," "serene," "moody," "energetic," "melancholic," "mystical." You can also specify camera details: "shot on 35mm film," "shot on a Sony A7R with a 50mm lens," "IMAX cinematography."

An example prompt built this way:

A cyberpunk street market in Tokyo at night, neon signs reflecting in wet pavement, crowds of people in leather jackets and holographic clothing, food stalls selling glowing ramen, dramatic backlighting, shallow depth of field, cinematic, shot on Kodak film, highly detailed, intricate.

This prompt takes 35 words and provides enough specific direction that most generation attempts will produce visually coherent, sophisticated images that read as intentional rather than accidental.

When you are learning, shorter prompts are better. A five-word prompt like "snowy mountains at dawn" is easier to iterate on than a 100-word prompt because you can adjust one variable at a time. Once you understand how Midjourney interprets your language, longer prompts become more controllable.

Source: Midjourney Community Handbook (2026) — synthesized from thousands of shared prompts in the community channels and documented techniques in official Discord resources.

Parameters: Aspect Ratio, Quality, Style, and Variants

After your prompt text, Midjourney accepts parameters, which are flags that modify how the image is generated. Parameters begin with -- and appear at the end of the command. Understanding these is essential for professional control.

--ar (aspect ratio) sets the width-to-height ratio of the output image. Midjourney's default is 1:1 (square). Common ratios are --ar 16:9 for widescreen (useful for backgrounds and cinematic images), --ar 9:16 for portrait (useful for character illustrations and phone wallpapers), --ar 4:3 for traditional photography, and --ar 3:2 for book covers. If you do not specify --ar, all outputs are square.

--quality or --q controls the detail level and generation time. Options are --quality 0.25, --quality 0.5, --quality 1, or --quality 2. The default is --quality 1. Higher quality takes proportionally longer (0.25 is four times faster, 2 is twice as slow) and uses more GPU minutes from your monthly allowance. For quick concept sketches, --quality 0.25 is sufficient; for professional print work, --quality 2 is standard.

--niji switches the model from the standard Midjourney model (which is photorealistic) to the "niji" model, which is optimized for anime and anime-adjacent styles. If you are generating anime, manga, or cartoon illustrations, --niji produces dramatically better results than the standard model. You can also use --niji 6 to access the latest niji version.

--style allows you to apply preset stylistic modifications. Common options are --style raw (removes Midjourney's default "prettification") and --style anime (applies anime aesthetics). As of June 2026, available styles include "expressive," "raw," and version-specific variations.

--s (stylize) is a numerical parameter from 0 to 1000 that controls how much Midjourney "interprets" your prompt versus how literally it follows your text. Lower values (0–100) produce literal, adherent results; higher values (500–1000) produce more stylized, artistic interpretations. A value of --s 50 will follow your exact words closely; --s 950 will use your prompt as inspiration but add significant artistic flourish.

--chaos (0 to 100) controls the randomness and variety in each generation grid. Low chaos (10) produces similar variations; high chaos (90) produces diverse results. For exploring new visual directions, high chaos is useful; for refining a specific idea, low chaos produces more consistent results.

An example command with multiple parameters:

/imagine a portrait of a woman in a gilded frame, oil painting style, Renaissance, golden hour lighting --ar 1:1 --quality 2 --s 700 --niji --chaos 30

This command generates a high-quality, anime-styled Renaissance portrait with moderate stylization and relatively consistent variations.

Source: Midjourney Official Parameter Documentation (2026) — comprehensive list maintained in the official Discord server and midjourney.com knowledge base.

Iteration: Upscaling, Variations, and Remixing

Once Midjourney produces an initial grid of four images, you have three primary options: Upscale, Variation, and Remix.

Upscaling increases the resolution and detail of a single image from the grid. When you upscale, Midjourney generates additional detail and sharpness. Most upscales increase from the initial 512×512 pixel grid to 1024×1024 pixels, though higher resolutions can be extracted with additional processing. Upscaling takes roughly 20 seconds and consumes GPU minutes from your monthly allowance. After upscaling, you can also request additional upscaling with the "2x" or "4x" enlargement buttons, which scale the image without additional AI generation—these are computationally cheap and do not consume GPU minutes.

Variations (labeled "V1" through "V4") generate four new variations based on the selected image. When you click "V2," Midjourney generates four new images that share visual qualities with your chosen image but introduce randomness and creative exploration. Variations are useful when you like the direction of an image but want to see alternative interpretations. Each variation still consumes GPU minutes.

Remix mode (activated with --remix or a dedicated button) allows you to modify your prompt while keeping the composition and structure of a previous generation. This is extraordinarily useful for iterative refinement. For example, if you generated an image with prompt "cyberpunk city at night" and you want to try "cyberpunk city at dawn with pastel colors," Remix mode will preserve the spatial layout and composition while applying the new prompt. Remix mode is the closest equivalent to "edit" functionality within Midjourney.

A typical iteration workflow looks like this: (1) Generate initial grid with broad prompt. (2) Identify one image in the grid that is close to your vision. (3) Request variations to explore nearby visual space. (4) Upscale one of the variations if it is satisfactory. (5) If not satisfied, use Remix mode to adjust the prompt and preserve composition.

Most professional work requires three to seven generations before achieving a result suitable for presentation or further refinement in external tools like Photoshop or Procreate. Iteration is not a sign of failure; it is the normal workflow, and experienced users budget 5–10 minutes per usable image.

Source: Midjourney Official Guide — upscaling, variation, and remix mechanics documented with step-by-step instructions on Discord.

Styles, Models, and Advanced Techniques

Midjourney's default model is powerful, but the platform also offers specialized models and style presets for specific use cases.

The anime model (niji) is trained specifically on anime and animation data, producing results that align with Japanese animation aesthetics. If you are generating character art, manga panels, or anything stylistically adjacent to animation, --niji 6 consistently outperforms the standard model.

The photo mode is tuned toward photorealistic output. While the standard model is reasonably photorealistic, adding "photograph" or specifying camera settings ("shot on a Hasselblad," "Kodak Portra 400") and film types ("35mm," "medium format") explicitly pushes toward photography aesthetics.

Negative prompts allow you to specify what you do not want. For example, if your prompt is "fantasy warrior," you might add --no blur, watermark, distorted hands to exclude common AI failure modes. Negative prompting is a skill: being too specific or too broad with negatives can paradoxically worsen results, so most professionals use negative prompts sparingly, only for obvious exclusions.

Seed (activated with --seed [number]) allows you to reproduce the exact same random state across multiple generations. This is useful for subtle variations: generate an image with --seed 12345, then request variations on that seed, and Midjourney will produce related images that branch from the same starting point. Without specified seeds, every generation includes new randomness.

Reference images (using --iw or image weight, combined with image URLs) allow you to incorporate visual reference. You paste an image URL into your prompt, and Midjourney uses that image as a style or composition guide. This is powerful for maintaining visual consistency across character designs or matching a specific photographic style.

One advanced technique is prompt weighting, which uses colons to emphasize words: subject::3 style::2 mood::1 tells Midjourney to prioritize subject at 3x importance, style at 2x, and mood at 1x. Weighted prompts require practice but produce remarkably controlled results for professionals who refine them iteratively.

Source: Midjourney Advanced Features Documentation (2026) — techniques compiled from official updates and professional user community shared practices.

Common Mistakes and How to Avoid Them

Even experienced users encounter predictable failure modes. Understanding these prevents wasted GPU credits and frustration.

Hands and fingers are a notorious weak point for all text-to-image AI models, including Midjourney. If hands are critical to your image, do one of the following: (1) obscure hands (folded behind back, in pockets, out of frame), (2) use extreme zoom on face or upper body, (3) use negative prompt --no hands and regenerate with focus on gesture rather than detail, or (4) upscale to high resolution and manually edit hands in Photoshop using reference. Never expect Midjourney to generate anatomically correct hands on the first try; plan for post-processing.

Repeated objects and counting are difficult. If you want a woman holding three apples, Midjourney often produces two, four, or apples distributed oddly. For precise object counts, descriptions like "woman holding apples" without a number, or reformulation as "woman with apples scattered around her," produce more reliable results than explicit counts.

Text within images should generally be avoided or treated as a post-processing task. Midjourney struggles with legible, intentional text, so if your image requires readable signage or logos, generate the image without text and add it afterward in design software.

Over-specification is a subtle mistake: a 200-word prompt that specifies every color, texture, and composition detail often produces worse results than a 30-word prompt that trusts Midjourney's aesthetic judgment. The AI interprets contradictory signals from over-specified prompts, resulting in visual chaos. Aim for clarity, not completeness.

Aspect ratio mismatches occur when you generate a tall portrait with --ar 16:9 (widescreen). Plan your aspect ratio before generating so that the composition fits the format. Regenerating with the correct aspect ratio after seeing a widescreen composition squash or stretch awkwardly is a frequent waste of time.

Source: Midjourney Community Best Practices (2026) — compiled from troubleshooting threads and shared professional workflows in Discord channels.

Ethics, Copyright, and the Current Limits of AI Art

Midjourney's terms of service state that users own the copyright to images they generate if they have a paid subscription. However, the underlying training data raises ongoing questions about fairness to artists whose work trained the model without explicit consent. This is not a settled legal question, and artists' rights organizations continue to pursue litigation against AI image generators. Using Midjourney ethically requires understanding these tensions.

Copyright ownership of your generated images belongs to you if you hold a paid subscription, and you are free to commercialize, sell, or modify them. If you use the free trial tier, images are considered "public," and Midjourney retains the right to display them in galleries. Always use a paid account if generating for commercial purposes.

Training data ethics remain contested. Midjourney was trained on billions of images from the internet, including copyrighted artwork, without explicit permission from artists. Some artists have successfully opted out of future training iterations, and organizations like the Authors Guild have filed lawsuits. As a user, you should be aware that your use of Midjourney indirectly supports this training pipeline, and some artists view AI image generators as a threat to their livelihoods. Consider disclosing when work is AI-generated, especially in professional or public contexts.

Deepfakes and misuse are potential harms. Midjourney's terms explicitly prohibit generating images intended to deceive (photorealistic images designed to look like real events or people without disclosure), illegal activity, or abuse. Midjourney monitors usage and has terminated accounts for violations. For professional work, assume your usage is logged and auditable.

Fair attribution is the emerging standard: if you generate an image using Midjourney and present it publicly, disclosing "Generated with Midjourney" is becoming expected practice, especially in academic or journalistic contexts. In commercial illustration and advertising, disclosure is less standardized, but transparency is increasingly valued by audiences.

The technology is improving rapidly. As of June 2026, Midjourney produces photorealistic results that can fool observers, and hands and anatomical details are far more reliable than they were in 2022. The human role remains essential: prompting, iteration, refinement, integration with design, and ethical judgment. Midjourney is a tool for amplifying human creativity, not replacing it. Understanding its limitations and using it conscientiously is the professional standard.

Source: Midjourney Terms of Service (2026) — copyright ownership, prohibited uses, and user rights documented officially.

Source: Authors Guild Complaint (2023), Authors Guild v. OpenAI et al. — ongoing litigation regarding training data copyright and artist consent, relevant to all generative AI platforms.

Frequently Asked Questions

Q: How much does Midjourney cost, and which tier should I choose?
A: Midjourney's Basic plan costs $10/month (200 fast GPU minutes), Standard is $30/month (15 hours), and Pro is $120/month (unlimited). For hobbyists generating a few images per week, Basic is sufficient. For professionals or daily users, Standard or Pro is necessary because CPU minutes disappear quickly with high-quality iterations.

Q: Can I use Midjourney images commercially or for selling products?
A: Yes, if you have a paid subscription. Midjourney states that you own the copyright to generated images on paid accounts and may sell, license, or modify them. Free trial images are not owned by you; Midjourney retains public rights.

Q: Why do my images look generic compared to professional examples?
A: Generic output usually results from vague prompts or insufficient style direction. Include specific style references ("oil painting by Caravaggio," "concept art by John Singer Sargent"), lighting details ("golden hour, rim lighting"), and mood ("dramatic," "serene"). Most generic results improve dramatically with 30–50 word prompts instead of three-word prompts.

Q: What do I do if an image has distorted hands or anatomically incorrect features?
A: Hands are a known limitation. Options are: (1) regenerate with hands obscured (behind back, in pockets, out of frame), (2) use negative prompt --no distorted hands and regenerate, or (3) upscale the image to high resolution and manually edit hands in Photoshop using reference images. Post-processing is normal for professional work.

Q: How do I make images more consistent when generating a series of images (like character designs)?
A: Use --seed [number] to lock the random state, and regenerate with slight prompt variations. Alternatively, use reference images with --iw (image weight) to guide style consistency. Remix mode also preserves composition while updating the prompt.

Q: Is it ethical to use Midjourney commercially? Are there copyright concerns I should know?
A: Ethically, yes, if you disclose the use of AI generation—especially in journalistic, academic, or artistic contexts where transparency is valued. Legally, you own the copyright to your generated images (paid subscription). However, the training data raises contested questions about artist consent; some artists view AI generators as unfair. Consider your audience's expectations around disclosure.

Recommended Reading

Read next:

• Text-to-Image AI: DALL-E, Midjourney, and Stable Diffusion Explained
• AI Art Copyright and Ethics: Ownership, Training Data, Liability
• Generative AI Art: How Diffusion Models Work

For deeper context on AI capabilities and limitations, see What Is Generative AI and How Is It Changing Everything.