From Text to Art: Everything You Need to Know About Using Midjourney

ThatPainter is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More

Midjourney turns text and image references into visual concepts, illustrations, photographs, and design starting points. Its best results rarely come from one perfect prompt. They come from directing a scene, choosing the right controls and references, comparing variations, editing the strongest result, and finishing details elsewhere when necessary.

In 2026, Midjourney works through both its website and Discord. The labels, models, and available features change frequently, so treat version-specific instructions as temporary and use the model and settings shown in your account.

What is Midjourney?

Midjourney is a generative image platform. You describe an image in natural language, optionally provide reference images, and the system produces new visual interpretations. Its current services also expose image-editing and video features.

It is not a conventional drawing application or a deterministic renderer. A prompt is guidance, not a precise design specification. Midjourney may change faces, hands, typography, object details, composition, or identity even when references are supplied.

Its main creative tools include:

• Text-to-image: generate images from a written prompt.
• Image prompts: influence composition, content, colors, and visual direction with one or more images.
• Style references: guide visual treatment without making the reference the main subject.
• Omni-reference: guide the inclusion of a character, object, vehicle, or creature.
• Personalization and moodboards: steer results toward preferred or shared aesthetics.
• Describe: analyze an image and suggest starting prompts.
• Editor: expand, crop, repaint, replace, or retexture parts of an image.

See Midjourney’s official documentation for current model and feature availability.

Website or Discord?

Using the website

1. Go to Midjourney and sign in.
2. Open Create.
3. Enter a prompt in the prompt bar.
4. Confirm the model, aspect ratio, speed, and other settings.
5. Generate an image, inspect the result, and iterate.

The website also provides Explore for discovery, along with personalization, moodboards, reference tools, editing, and newer image or video controls. Exact labels may change.

Using Discord

1. Join the official Midjourney Discord, or message the Midjourney Bot directly.
2. Type /imagine.
3. Enter your prompt and submit it.
4. Use the returned controls for variations, upscaling, or further changes.

Website and Discord settings can be synchronized. Discord’s current settings documentation includes model selection, RAW mode, stylization, personalization, public or Stealth mode, Remix, variation modes, and GPU speed controls. Check the current settings guide before relying on a particular command.

Midjourney’s models are moving targets

Model names and capabilities can change quickly. Midjourney’s help center currently references V8.1, while a separate official interface page displayed V8.2 during the same research period. Do not assume that one model is universally current across the website, Discord, and documentation.

Choose the newest generally available model shown in your account, and record the model and date for work you may need to reproduce. A prompt or parameter that works in one model may behave differently in another. Current availability can be checked in the Help Center and Midjourney interface.

How to write a useful prompt

A practical prompt usually combines:

Subject + action or state + setting + composition + visual treatment + lighting and color + technical constraints

×

Quick wins for a faster PC:

Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

A small red fox crossing a snow-covered forest path at dawn, viewed from a low three-quarter angle, mist between the trees, soft blue shadows and warm rim light, cinematic naturalism --ar 16:9

• Subject: what should appear?
• Action or state: what is happening?
• Setting: where is the scene?
• Composition: close-up, overhead, centered, asymmetrical, or with negative space?
• Visual treatment: editorial photograph, ink illustration, clay sculpture, or 3D render?
• Lighting and color: candlelight, hard sun, muted earth tones, or saturated primaries?
• Constraints: aspect ratio, model, stylization, RAW mode, or references.

Put important visual information early. Describe what should be visible rather than explaining an abstract concept. “An anxious astronaut sitting alone in a laundromat” is more useful than “a visual meditation on alienation.” Likewise, “large empty area on the right” is more actionable than “make it suitable for text.”

More adjectives do not automatically mean more control. Contradictory instructions, generic style piles, and unrelated concepts make results less predictable. Camera and lens terms can suggest photographic treatment, but they are aesthetic guidance—not guaranteed technical metadata.

Negative prompts

Use a small number of meaningful exclusions with --no:

A clean product photo of a blue ceramic mug on a pale stone table, soft studio light --no text, logo, watermark

Exclusions are not guarantees. A long list can introduce unintended interactions, so use them sparingly.

Controls that materially change the result

Control	What it does	Useful starting point
--ar	Changes the image shape.	Choose the final use before generating.
--stylize or --s	Controls the strength of Midjourney’s aesthetic interpretation.	Lower for literal briefs; higher for expressive exploration.
--raw	Reduces some automatic stylistic influence.	Try it for photography or controlled composition.
--chaos or --c	Controls how different initial results are from one another.	Low for consistency; higher for brainstorming.
Variety	Encourages more variation in possibilities.	Useful during early exploration.
Weirdness	Pushes results away from ordinary interpretations.	Use for unusual concepts, not reliability.
GPU speed	Balances generation speed against available Fast or Relax time.	Use Fast for deadlines and Relax for volume when available.

Aspect ratio

Use --ar 1:1 for square posts or avatars, --ar 16:9 for presentations and landscape scenes, --ar 9:16 for phone screens, and --ar 4:5 for portrait social imagery.

Aspect ratio is not final print resolution. You may still need cropping, upscaling, or layout work.

Stylization and RAW

Midjourney’s Discord settings list stylization presets of 50, 100, 250, and 750, with 100 as the default in that interface. Lower values generally favor literal instructions; higher values encourage stronger artistic interpretation and may sacrifice small details.

A minimal black-and-white editorial portrait --s 50

A red bicycle leaning against a concrete wall, overcast daylight --raw

Chaos, Variety, and Weirdness

Use lower chaos when the initial options should share a close interpretation. Increase it for divergent ideas. Variety and Weirdness are current web controls, but their labels and behavior may differ across models and Discord. Neither guarantees originality, coherence, or usefulness.

Choosing the right reference tool

Your goal	Best starting tool
Influence composition, colors, or content broadly	Image prompt
Carry over visual treatment	Style reference
Keep a character or object recognizable	Omni-reference
Build a recurring personal aesthetic	Personalization
Combine preferred looks	Moodboard
Modify an existing image	Editor
Turn an image into prompt ideas	Describe

Image prompts

Image prompts provide broad visual guidance; they do not normally reproduce the source exactly.

IMAGE_URL A_IMAGE_URL B_IMAGE_URL
A futuristic greenhouse on a rooftop at sunset --ar 16:9

On the website, drag an image into the relevant prompt or reference area. Upload only images you have permission to use.

Style references: --sref

A glass observatory above the clouds --sref IMAGE_URL

A style reference may influence color relationships, texture, lighting, and rendering approach. It does not guarantee exact subject copying, exact layout, consistent branding, or permission to imitate a living artist. The V7 announcement also documents --sref random and compatibility considerations for older style codes; see Midjourney’s guide.

Personalization and moodboards

Personalization adapts results toward a user’s preferences; moodboards collect visual direction for a project or shared aesthetic. Personalization is not a fixed preset: it can change as preferences are rated or liked, and profiles may differ between accounts.

Use it for a recognizable personal style or open-ended exploration. Avoid it when a client requires neutral, strictly reproducible output. Depending on the account, Discord syntax may include --p. Current web personalization changes are documented here.

Omni-reference: --oref

Omni-reference is intended to place a particular character, object, vehicle, or non-human creature into a new scene. In the web workflow, select a compatible model, drag an image into the Omni-reference area, adjust reference strength, and generate. In Discord:

A robot mechanic repairing a motorcycle in a rainy garage --oref IMAGE_URL --ow 100

--ow controls reference strength, but useful values and compatibility are model-specific. Clothing, age, expression, proportions, and materials may still change. A clean reference with one dominant subject usually reduces ambiguity. See the official Omni-reference announcement.

Describe

Describe works in the opposite direction: right-click an image or drag it into Describe, and Midjourney proposes four text prompts. It is useful for learning descriptive vocabulary or creating a starting point, but the results are interpretations—not authoritative captions or reverse-engineered copies. Details are in Midjourney’s web announcement.

A reliable iteration workflow

1. Generate a broad first pass.
2. Identify the strongest composition rather than chasing perfection immediately.
3. Create variations or upscale the promising result.
4. Change one major variable at a time.
5. Use Remix when you want to change the prompt while preserving a selected result.
6. Use the Editor for expansion or localized changes.
7. Upscale only after composition and subject identity are acceptable.
8. Finish typography, layout, retouching, and color correction elsewhere.

For example, begin with:

A modern bookstore inside a renovated train station --ar 16:9

Then establish composition and treatment:

A modern independent bookstore inside a renovated 1920s train station, long central aisle, tall arched windows, warm afternoon light, shelves clearly visible, editorial interior photography --ar 16:9 --raw

Finally change only the mood:

A modern independent bookstore inside a renovated 1920s train station, long central aisle, tall arched windows, shelves clearly visible, rainy blue-hour light, editorial interior photography --ar 16:9 --raw

This controlled loop is more reliable than randomly adding adjectives.

The Editor and retexturing

Midjourney’s Editor can upload, expand, crop, repaint, replace, or retexture an image while attempting to preserve its underlying scene structure. It is useful for adapting a composition to a new ratio, replacing a localized object, or changing materials and color.

It is not equivalent to layer-based Photoshop editing. Masks do not always constrain changes perfectly, and repeated edits can cause faces and object geometry to drift. Text and logos remain unreliable. See the external Editor announcement.

What Midjourney still struggles with

Text and logos

Text rendering has improved, but spelling, long copy, legal wording, labels, and logos still require inspection. For professional work, generate the visual with blank signage where possible, try the wording if necessary, proofread every letter, and typeset final copy in a design application.

Hands and small details

Hands, jewelry, cables, utensils, complex machinery, and repeated small objects can remain inconsistent. Try a simpler composition, closer crop, lower stylization, RAW mode, or references, then retouch externally.

Character consistency

Omni-reference can improve continuity but does not create a guaranteed character model. Large projects such as books, games, and campaigns may require a carefully chosen base image, repeated references, manual selection, editing, and correction.

Exact reproduction and factual accuracy

Midjourney creates new interpretations rather than pixel-perfect copies. Do not rely on generated images for exact technical diagrams, factual evidence, safety instructions, medical information, pricing, or legal text without independent verification.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

Pricing, commercial use, and privacy

Official plan prices checked August 18, 2026 were:

Plan	Monthly	Annual	Fast GPU time	Relax	Stealth
Basic	$10	$96/year	3.3 hours / 200 minutes	No image Relax	No
Standard	$30	$288/year	15 hours	Unlimited images	No
Pro	$60	$576/year	30 hours	Unlimited images and SD video	Yes
Mega	$120	$1,152/year	60 hours	Unlimited images and SD video	Yes

Annual billing is described as a 20% upfront discount. Extra Fast time is listed at $4 per hour. Plans automatically renew; unused Fast time does not carry over, and cancellation generally does not refund the current subscription period. Confirm current details on the official plan page before subscribing.

Which plan fits?

• Basic: occasional experimentation and learning.
• Standard: the practical default for regular image creators who want unlimited image Relax mode.
• Pro: privacy-sensitive professionals, heavier users, and businesses that need Stealth or meet certain commercial requirements.
• Mega: high-volume production that regularly exhausts Pro capacity.

Relax is available for images on Standard, Pro, and Mega. Relax video is limited to Pro and Mega, and queue times vary with demand. The help center currently says Turbo is not supported in V8.1. “Unlimited” means the specified Relax mode remains subject to queueing and rate limits.

Commercial use and ownership

Midjourney’s commercial-use guidance says subscribers own images and videos they create, including after cancellation, to the fullest extent possible under applicable law and subject to its Terms. A business or employee of a company earning more than $1 million in annual gross revenue must use a Pro or Mega plan to own or commercially use its assets under those terms.

This is contractual usage permission, not a guarantee of copyright protection. Copyright rules differ by jurisdiction. You must also have rights to uploaded images, faces, trademarks, client assets, and artwork. Upscaling another user’s image does not make you its owner. Midjourney’s Terms also grant the company a broad license over user inputs and generated assets. Read the commercial-use guidance and Terms of Service.

Privacy

Stealth Mode is available only on Pro and Mega and affects visibility on Midjourney’s website. It does not hide creations made in public Discord channels from people in those channels. For confidential work, use the website Create page, Discord direct messages, or a private Discord server, together with Stealth where appropriate. Making an image private is not the same as deleting it. See Midjourney’s Stealth documentation.

Troubleshooting

The result ignored the important subject

Shorten the prompt, put the subject and action first, lower stylization, try RAW mode, or add an image or Omni-reference. Change one variable at a time.

The style is right but the composition is wrong

State the framing, camera angle, subject position, and negative space. Use an image prompt for layout, select a variation from the strongest composition, or use the Editor for expansion.

The character keeps changing

Use a clean single-subject reference, compatible Omni-reference settings, stable clothing and camera distance, and a strong base image. Expect manual correction.

A parameter does nothing

Check its spelling, position at the end of the prompt, model compatibility, and whether the website exposes the same control. Parameters and ranges are not universal across versions.

Generation is slow

Check whether you are using Fast or Relax, your remaining GPU time, and current queue demand. Standard, Pro, and Mega offer image Relax; higher plans provide more Fast time.

When Midjourney may not be the right tool

Consider another workflow if you need guaranteed typography, pixel-perfect editing, local or offline generation, enterprise administration, or a character that remains identical across a large production.

Adobe Firefly is worth considering for Adobe users who prioritize an integrated design and editing workflow. OpenAI image generation may suit conversational or API-based workflows. The Stable Image ecosystem may appeal to users seeking model flexibility or developer control, though setup varies.

Best practices for repeatable projects

• Save the complete prompt and parameters.
• Record the model, personalization profile, references, and date.
• Keep selected job IDs and source images.
• Separate content decisions from style decisions.
• Proofread every visible word.
• Secure permissions for uploaded references.
• Plan external typography, retouching, compositing, accessibility, and final-size checks.

Midjourney is most effective when treated as a visual ideation and production stage—not as a magic prompt box or a complete replacement for art direction and finishing work.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.