Midjourney turns text prompts into four image variations at a time, then lets you steer the look with style references, personalization profiles and parameters — and animate any still into a five-second video, on the web or inside Discord.
Used this tool? Rate it
Used this tool? Rate it
Midjourney is a generative image and video service that turns written descriptions into finished visuals. You type a sentence describing what you want to see, and the model returns a set of four images built from that description. From there the work becomes iterative: you pick the frame closest to your intent, produce variations of it, enlarge it, repaint a region, extend the canvas, or animate it into a short video clip. The official documentation frames the product plainly as a portal for transforming text into images and videos, and that phrase captures the scope accurately — this is not a photo retoucher that happens to include AI features, but a generation engine whose editing tools all operate by regenerating pixels rather than pushing them around.
The service occupies a specific position in a crowded field. Where some competitors compete on raw controllability or on integration into an existing design suite, Midjourney has consistently competed on aesthetics: the default output tends toward dramatic lighting, confident composition, and a strong point of view, sometimes at the cost of literal prompt obedience. Recent model releases make this explicit. The V8.2 model became the default on July 24, 2026, and the documentation describes it as an update focused on aesthetics, image quality, and Personalization, producing images that are more creative, bold, sophisticated, and edgy. That house style is the product's main draw and, for certain jobs, its main constraint.
Two things distinguish the day-to-day experience from most rivals. First, access is sold as GPU time rather than as a credit count per image: subscribing effectively buys a monthly allowance of processing minutes, and the documentation compares GPU time to the data allowance on a mobile phone plan. Second, the service runs in two places at once — a full web application at midjourney.com and a bot inside Discord — with the same underlying models and the same subscription covering both. Typical users are illustrators, concept artists, art directors, marketers producing campaign visuals, game and film pre-production teams building mood boards, and independent creators who need a high volume of original imagery without commissioning it.
Text-to-image generation in grids of four: A single prompt returns four candidate images rather than one, which turns generation into a selection exercise instead of a lottery. The official getting-started flow describes this directly: after pressing Enter, Midjourney creates a set of four images that you watch take form in real time. Working in fours makes it far cheaper to discover an unexpected direction than to specify it in advance.
Style References and style weight: The --sref parameter lets you point the model at an existing image whose look and feel you want to match, while --sw controls how strongly that reference pulls the result. Combined with Style Reference Codes and the Style Creator, this is the mechanism for holding a consistent visual language across a whole campaign or series rather than re-describing the aesthetic in prose every time.
Omni Reference for consistent subjects: Where Style References carry the look, Omni Reference (--oref) carries the subject — a person's likeness or the form of a specific object — so the same character or product can appear across many images. It replaced the older Character Reference approach from V7 onward, and is the practical answer to the hardest problem in production work: keeping one protagonist recognizable across a sequence.
Personalization profiles and moodboards: The service can learn your individual taste from ratings you supply and from moodboards you assemble, then bias generations toward it via the --p parameter. The V8.2 release notes state that Personalization understands your aesthetic tastes much better, especially on profiles with a large amount of points, and that profiles built for V8.2 draw on a much larger and improved pool of images.
Image-to-video animation: Any still — one Midjourney generated or one you upload — can become a five-second video clip, with the original image acting as the first frame and a text prompt optionally steering the motion. Motion strength is selectable through --motion low and --motion high, videos can be made to loop with --loop, and a custom final frame can be set with --end.
Generative editing tools: Beyond first generation, the toolkit includes Variations, Upscalers, Pan for extending the frame directionally, Zoom Out for pulling the camera back, Remix for altering a prompt mid-stream, and an Editor with region-level repainting. These let a nearly-right image be finished in place instead of being regenerated from scratch and losing what already worked.
A parameter language for fine control: Prompts accept a documented set of switches appended at the end — aspect ratio (--ar), stylization strength (--stylize), unconventionality (--weird), variety (--chaos), exclusions (--no), seamless tiling (--tile), reproducible seeds (--seed), and Raw mode to strip default styling. The syntax is strict: parameters go after the prompt text, need a space before the dashes, and must not contain punctuation.
Concept art and pre-production: Art departments in games, film, and animation need to explore a visual space quickly before committing to expensive production. Generating four variants per prompt and then locking a direction with Style References compresses what used to be days of thumbnail sketching into an afternoon, and Omni Reference keeps a chosen character consistent as the exploration widens into a full sequence of shots.
Marketing and campaign visuals: Social posts, display ads, landing page headers, and seasonal campaign art all demand a steady stream of on-brand imagery. A team can establish a look once, save it as a style reference or personalization profile, and then produce dozens of variations that share a visual identity — avoiding both stock-photo genericness and the cost of a full photoshoot for every campaign beat.
Editorial and publishing illustration: Blogs, newsletters, reports, and book projects frequently need illustration for abstract subjects where no photograph exists. Because generation is prompt-driven, an editor can commission a metaphorical image for a difficult concept in minutes, iterate on tone through stylization settings, and settle on something specific to the piece rather than reaching for a stock image everyone else has used.
Product and packaging exploration: Industrial designers and brand teams use the model to visualize product forms, materials, colorways, and packaging in context before any prototype exists. Omni Reference makes it possible to hold one product form steady while varying environment and lighting, which is useful for testing how a design reads across different presentation contexts.
Short motion assets for social channels: Five-second clips animated from a strong still fill a real gap between static posts and full video production. Because the animation starts from an image already approved, the motion step inherits the composition and art direction decisions instead of relitigating them, and looping output suits the autoplay conventions of social feeds.
Personal creative practice and style development: Independent artists use the tool as a sketchbook for visual ideas — testing compositions, developing a recognizable personal aesthetic through Personalization, and building a body of reference material. The community dimension matters here too, since public generations are visible and remixable by default, which turns the gallery into a live source of technique.
Subscribe before you begin. Unlike many competitors there is no standing free tier to explore first; the official getting-started guide places subscribing ahead of creating, so choosing a plan is genuinely step one rather than a later upsell.
Log in at midjourney.com and open the Create page. This is the main workspace, where generations appear in real time as they render. Users who came in through Discord should sign in with the Continue with Discord option so that the two surfaces share one account and one gallery.
Write your prompt in the Imagine bar. The bar sits at the top of the Create page; type a description of what you want to see and press Enter. Midjourney creates a set of four images, and once the progress reaches 100% the grid is complete and ready to act on.
Add references and parameters to steer the result. Attach a style reference to match an existing look, an Omni Reference to carry a subject across images, or append parameters such as --ar for aspect ratio and --stylize for artistic intensity. Remember that parameters belong at the very end of the prompt with a space before the dashes and no punctuation inside them.
Refine the image you like. Use Variations to explore near neighbors of a promising frame, Upscale to increase resolution, Pan and Zoom Out to extend the composition, and the Editor to repaint a specific region. This is where most of the real work happens — the first grid is a starting point, not a deliverable.
Animate or export. To turn a still into motion, open the image and use the Animate buttons, choosing low or high motion; to animate your own picture, add it as a starting frame from the image panel. Finished work is downloaded, sorted, and filed from the Organize page, which is also where folders and filters live.
Describe the image, not the request. Prompts work best as descriptions of a finished picture — subject, setting, lighting, camera treatment, medium — rather than as instructions addressed to an assistant. Phrases like "please make me" spend tokens without adding visual information.
Change one variable at a time. When a result is close but wrong, resist rewriting the whole prompt. Adjust stylization, swap the style reference, or nudge the aspect ratio while holding everything else constant, so you learn which lever actually moved the output.
Lock a seed when comparing options. Reproducible seeds let you test whether a wording change genuinely improved the image or whether you simply got a luckier roll. Without this discipline, prompt experimentation degrades into superstition.
Build a personalization profile early, then trust it. Rating images is tedious for an hour and valuable for months. Since the current model's personalization improves noticeably on profiles carrying a large amount of rating data, front-loading that effort pays back across every subsequent project.
Separate look from subject. Use Style References for aesthetics and Omni Reference for the thing being depicted. Trying to force both through prose in a single prompt is the most common reason character consistency collapses across a series.
Budget your fast time deliberately. Processing one image prompt usually takes about one minute of GPU time and one video prompt about eight minutes, so exploratory work is cheap and video is not. Draft and Relax modes exist precisely so that wide exploration does not consume the allowance reserved for finals.
Reduce batch size when animating. Video prompts default to four clips, but batch size can be set to one or two. On a plan with limited fast hours, this single setting is the difference between finishing a project and running dry mid-month.
Expect moderation friction and plan around it. The official guidance is candid that seemingly innocent prompts may be blocked by the filters, and notes that blocked jobs do not cost credits. Rephrasing is usually faster than appealing.
Concept artists and illustrators: Professionals who need to generate and evaluate many visual directions quickly, and who value a distinctive aesthetic baseline over literal-minded rendering of every prompt clause.
Art directors and brand designers: Teams responsible for visual consistency across many deliverables, who benefit most from style references, reusable codes, and personalization profiles that encode a house look.
Marketing and social media teams: Groups producing continuous campaign imagery on deadline, where the alternative is stock photography or a production budget that does not exist.
Game, film, and animation pre-production: Departments building mood boards, environment studies, and character explorations before committing to expensive asset creation.
Independent creators and small studios: Solo operators who need professional-grade visuals without a photographer, illustrator, or retoucher on call, and who can absorb the learning curve in exchange for output quality.
Product and packaging designers: Practitioners visualizing forms, finishes, and shelf presence early, when physical prototyping would be premature.
Educators and researchers working on visual culture: Users who need to demonstrate, examine, or critique generative imagery directly, including its failure modes and its licensing complications.
Web application: The primary surface is the browser app at midjourney.com, containing the Create page with its Imagine bar and settings panel, the Organize page for sorting and downloading, plus Personalization, moodboards, folders, and the Editor.
Discord: The original interface remains fully supported. Generations run through commands in Discord channels or direct messages, with settings adjusted via the settings command and grid buttons for upscales and variations. The official documentation maintains parallel On Web and In Discord instructions for major features, and a single subscription covers both.
Cross-surface account linkage: Accounts connect across the two environments, so images generated in one appear in the same gallery as the other, provided you sign in with the matching login method.
Mobile access: The web application is reachable from a mobile browser, and the documentation references pressing the send button when using a mobile device. Availability of dedicated native applications should be confirmed on the official site rather than assumed. (To be verified)
No public generation API: Programmatic access is not part of the documented product surface; workflows are built around the web app and Discord rather than around API calls. (Based on public information)
There is no free tier. The official getting-started path requires choosing and subscribing to a plan before creating anything, so evaluation happens after payment rather than before it — a meaningful difference from competitors that offer limited free generations. Access is structured as four subscription tiers: Basic, Standard, Pro, and Mega. All plans are subscriptions that renew monthly or yearly unless cancelled, and committing to an annual plan carries a 20% discount with the full year paid upfront. As of the documentation available in August 2026, the published monthly prices are $10, $30, $60, and $120 respectively, with annual pricing that works out to roughly $8, $24, $48, and $96 per month.
What separates the tiers is mostly throughput and privacy rather than model access. Each plan includes a monthly allotment of Fast GPU time — 3.3 hours on Basic, rising to 15, 30, and 60 hours on the higher tiers — and unused Fast time does not carry over when the subscription renews. Relax Mode, which queues jobs without consuming the Fast allowance, begins at the Standard plan for unlimited images; unlimited video generation in Relax Mode requires Pro or Mega. Stealth Mode, which keeps creations private, is available only on the Pro and Mega plans. Additional Fast time can be purchased at a published rate of $4 per hour, and the plans also differ in how many prompts can run concurrently. Commercial use carries its own threshold: businesses grossing more than $1,000,000 USD a year need a Pro or Mega plan to use outputs commercially. Exact entitlements should always be confirmed against the current official pricing documentation, since tiers and allowances have changed repeatedly.
Adobe Firefly: Positioned around commercial safety and suite integration, with training data intended to reduce rights ambiguity and direct connections into Photoshop and Illustrator workflows. Chosen most often by teams whose legal review is stricter than their aesthetic ambition.
Stable Diffusion and its ecosystem: Open models that can run locally or on your own infrastructure, with community fine-tunes, controllable conditioning, and no per-image service cost. It trades polish and convenience for control, privacy, and extensibility.
DALL·E within ChatGPT: Strong at literal prompt following and conversational refinement, which suits users who want to describe adjustments in dialogue rather than learn a parameter syntax, though the default aesthetic is less opinionated.
Ideogram: Notable for text rendering inside images, which makes it a common companion tool for posters, logos, and any composition where legible typography inside the generated image matters.
Runway and other video-first tools: Built for motion from the start, with longer durations, timeline editing, and camera control beyond what a five-second image-to-video animation provides. Complementary rather than directly competing for still-image work.
No free evaluation path: Because subscription precedes creation, you cannot assess whether the aesthetic suits your project before paying. Users on public review platforms frequently frame this as the main friction in adopting the tool.
Precise control remains difficult: Reviewers on platforms such as G2, while rating image quality highly, consistently report frustration with achieving exact visual details and with errors in generated content that affect realism and consistency. The house style that makes output attractive also resists literal instruction.
Costs scale sharply with video: An image prompt costs roughly a minute of GPU time while a video prompt costs about eight, so motion work consumes an allowance far faster than still work. Independent reporting at the launch of the video model characterized video tasks as roughly eight times the cost of image tasks.
Default publicity of your work: Content is publicly viewable and remixable by default, and the terms describe Midjourney as an open community where others may use and remix content posted in a public setting. Privacy requires Stealth Mode, which is limited to the higher tiers — a real consideration for client work under NDA.
Unresolved copyright litigation: Major film studios have sued Midjourney over the use of copyrighted characters in training and output. Reporting through mid-2026 indicates the case remains in litigation and discovery, with Midjourney arguing that training on such material is permitted as fair use and no final ruling issued. Teams in regulated or rights-sensitive contexts should follow the matter rather than assume settled ground.
Ownership is broad but conditional: You own the assets you create, yet the terms carve out exceptions — upscaling another user's image leaves ownership with that original creator, and the revenue threshold changes which plan is required. Midjourney explicitly declines to advise on copyright, recommending a lawyer for jurisdiction-specific questions.
Moderation filters produce false positives: Prompts with no ill intent are sometimes blocked, which interrupts creative flow even though blocked jobs are not charged. External image uploads carry additional obligations, including holding the rights to what you upload and refraining from manipulating images of real people abusively.
No standing free tier is offered; the official onboarding requires selecting and subscribing to a plan before you can create. Anyone wanting to evaluate the aesthetic before committing generally has to rely on the large volume of public generations in the community gallery.
Generally yes on a paid plan — the documentation states you own the images and videos you create, and that ownership persists even if you later cancel your subscription. Two exceptions apply: upscaling someone else's image does not transfer ownership to you, and companies grossing over $1,000,000 USD annually must be on a Pro or Mega plan for commercial use.
It buys GPU time rather than a fixed number of images. Each plan includes a monthly allowance of Fast GPU hours that resets at renewal and does not roll over, with higher tiers adding unlimited Relax Mode generation, larger concurrency limits, and privacy features.
Processing one image prompt usually takes about one minute of GPU time, and one video prompt about eight minutes. A standard-definition set of four images costs roughly 0.8 minutes, an HD set roughly 1.3 minutes, and a batch of four standard-definition videos about eight minutes.
The default is whichever version Midjourney has most recently promoted, and it can be overridden per prompt with --v or set globally in the settings panel. V8.2 became the default on July 24, 2026, following V8.1, and older versions including V7 and V6 remain selectable.
Yes. A single image serves as the first frame and becomes a five-second clip, with optional text prompting and selectable low or high motion. Plan tier determines where this works: all tiers can generate video in Fast Mode, only Pro and Mega can do so in Relax Mode, and HD video requires Standard or above in Fast Mode.
Not by default. Creations are publicly viewable and remixable unless you use Stealth Mode, which is available only on the Pro and Mega plans. For confidential or client work, plan selection is effectively a privacy decision.
No. The full workflow runs in the browser at midjourney.com, including generation, editing, personalization, and organization. Discord remains supported for those who prefer it, and one subscription covers both surfaces.
Use Omni Reference, which carries a subject's likeness or an object's form between generations and replaced the earlier character reference approach. Pair it with a style reference so that the look and the subject are controlled by separate mechanisms rather than competing inside one prompt.
Major studios including Disney and Universal have brought copyright claims against Midjourney concerning training data and outputs depicting their characters. Coverage through mid-2026 describes the case as still in litigation and discovery, with Midjourney asserting fair use; no final ruling had been reported. Consult a qualified lawyer for advice specific to your situation.
