Midjourney teardown: 9.25M visits a month for the image generator with no free tier

Monthly visits

9.25M

Semrush, October 2026. Third-party estimate, not official MAU/DAU.

Image tools

Where the visits come from

United States28.61% (2.65M)
China6.62% (612.31K)
India4.24% (392.17K)

Midjourney draws an estimated 9.25M visits a month (Semrush, October 2026). That makes it the fifty-seventh largest site in our database of 195, just below InVideo at 9.37M and just above media.io at 9.15M, and the largest AI image generation tool we track, ahead of Leonardo.ai at 8.73M by only 0.52M. It is the fifteenth image or PDF tool we have torn down. The thing that sets it apart from the image tools around it is the combination of two facts: there is no free tier — the pricing page lists four paid plans and nothing free — and the company behind it is a self-funded lab of 60 people that has never taken venture capital. The operator is Midjourney, Inc. of San Francisco, founded by David Holz, who previously co-founded Leap Motion.

Data card: every figure below is sourced, and the source is named
SiteMidjourney, midjourney.com
What it isAI text-to-image generator: type a prompt and get four images back. Plus a V1 video model that animates a still image into a short clip, and a web editor for editing, panning, zooming, region variation and inpainting
Who it is forIllustrators, designers and creators who need high-quality AI-generated images
CategoryImage tools
Monthly visits9.25M, Semrush, October 2026. Third-party estimate, not official MAU/DAU
PricingNo free tier listed on the pricing page. Basic $10/month billed monthly or $8/month billed annually ($96/year); Standard $30/month or $24/month ($288/year); Pro $60/month or $48/month ($576/year); Mega $120/month or $96/month ($1,152/year). Annual billing saves 20% on every tier. Source: Midjourney pricing page, read 16 November 2026
Pricing modelSubscription only, metered by Fast GPU hours. Basic is capped at about 200 image jobs a month (about 800 images) instead of hours; Standard gets 15 Fast hours a month, Pro 30 and Mega 60. No free tier and no trial length stated on the pricing page
Country breakdownUnited States 28.61% (2.65M); China 6.62% (612.31K); India 4.24% (392.17K). Semrush, October 2026
OperatorMidjourney, Inc., San Francisco. Founded by David Holz, previously a co-founder of Leap Motion. Sources: Midjourney About page and Wikipedia, read 16 November 2026
Acquisition channelsNot publicly disclosed (see the acquisition section)

TL;DR

  • Midjourney draws an estimated 9.25M visits a month (Semrush, October 2026), the fifty-seventh largest site in our database and the largest AI image generation tool, ahead of Leonardo.ai at 8.73M by only 0.52M — a 5.6% gap. The United States is first at 28.61% (2.65M), China second at 6.62% (612.31K) and India third at 4.24% (392.17K). The top three together are 39.47%. The United States alone sends 2.65M visits, more than the whole traffic of CombinePDF at 2.5M, and India alone sends 392.17K visits, more than the whole traffic of Coggle at 362.99K. October fell 16.78% from 11.11M in September, and no public source states a reason for the drop.
  • The pricing is the story. There is no free tier: Basic costs $10 a month or $8 a month billed annually, Standard $30 or $24, Pro $60 or $48, and Mega $120 or $96, with annual billing saving 20% on every tier — $96, $288, $576 and $1,152 a year by arithmetic on the printed prices. Basic is capped at about 200 image jobs a month, Standard adds 15 Fast hours and unlimited Relax image generations, Pro adds 30 Fast hours and Stealth mode, Mega doubles that to 60 Fast hours. Video is included in the same subscription: SD video on Basic, HD from Standard up. Third-party reporting says free access was removed in March 2023, so the site has been subscription-only since — the only way in is $10 a month.
  • It survived by being the best-looking model and the most independent company in AI. The service entered open beta on 12 July 2022, and its founder told The Register the following month that the company was already profitable. Its own About page calls it a lab of 60 people and a lean, self-funded, distributed team, and it has never taken venture capital. Third-party trackers estimate revenue in the hundreds of millions a year, but they disagree with each other — roughly $300M to $500M for 2025 — and the company publishes no official figure, so this teardown does not print a single revenue number. The risk in the story is that the traffic figure comes from a single third-party source, the lawsuits over training data are still open, and the company publishes no acquisition mix, no conversion rate and no retention figure.

What problem does Midjourney solve, and for whom?

The problem is a blank page that needs a finished-looking image in minutes. An illustrator needs a concept to show a client before starting work, an art director needs a mood board, an architect needs a visual for the early stages of a project, and none of them wants to wait for a photoshoot or a stock search. Midjourney's answer is a prompt box: type a description, get four images back, pick one, upscale it. The master-table record puts the audience as illustrators, designers and creators who need high-quality AI-generated images, and the uses Wikipedia records fit that width — artists using it for rapid prototyping of concepts to show clients, advertising agencies using it to brainstorm and create original content, and architects using it for mood boards as an alternative to searching Google Images.

The site states its pitch in one line on its About page, read on 16 November 2026: "We're a lab of 60 people known for building the most beautiful AI models in the world." The pitch is beauty, not speed, not price and not breadth. The page says nothing about being the cheapest or the fastest, and the pricing page does not argue value either — it lists four tiers and the hours each one buys. That confidence is the product of being first to the look: Midjourney entered open beta on 12 July 2022, when AI images were still new, and the images it made looked better than the alternatives. The Economist used one for a magazine cover in June 2022, a Midjourney image won first place in the digital art competition at the 2022 Colorado State Fair, and a children's book illustrated with it was published in December 2022. All three are recorded as events, not as endorsements — the fair win and the book both drew public criticism from artists, which is part of the story too.

The audience is deliberately global and English-shaped at the same time. The United States is first at 28.61%, China second at 6.62% and India third at 4.24%, and the top three together are 39.47% — a flatter split than most tools in the database, where the top country often takes half the traffic. A product that sells in English prompts to anyone with a browser is expected to look like this. China in second place is worth noting: the company once banned images of Xi Jinping under its old moderation system, then moved to an AI-powered moderation system in May 2023 that allows a portrait but blocks depictions of leaders in situations such as arrest. That history is recorded in the risks section, without taking a position on it.

The product and its first screen

The public homepage is not a sales page. It is a short statement of intent — the About text quoted above, a paragraph about upcoming projects under the themes of imagination, coordination, reflection, beauty and human flourishing, and a careers note — because the product lives behind a login, in the web app and in Discord. The page as fetched on 16 November 2026 is about twenty lines long, which is itself a finding: a site with 9.25M visits a month has almost no public surface. There is no feature list, no pricing teaser and no gallery on the homepage. The product is the channel.

The product itself is two doors into the same model. The first is Discord, where it started: the Midjourney server launched on 14 March 2022, and users generate images with the /imagine command and a prompt, getting four images back that they can upscale. The second is the website, which launched in August 2024 alongside version 6.1 and consolidates the editing tools — image editing, panning, zooming, region variation and inpainting — into a single interface. Users once needed to have generated at least 1,000 images through the bot before they could use the website; that limitation has since been removed, per Wikipedia.

The model has been versioned in public, roughly every few months, and the version list is the product's release history: V1 in February 2022, V2 in April, V3 on 25 July, V4 on 5 November, V5 on 15 March 2023, V5.1 on 3 May, V5.2 on 22 June, V6 on 21 December, V6.1 on 31 July 2024, V7 on 4 April 2025, V8 on 17 March 2026, V8.1 on 14 April and V8.2 on 24 July 2026. V4 onwards was trained on Google TPUs. There is also Niji, a model tuned for anime, with Niji 7 released on 9 January 2026. The features added along the way are the ones that keep a professional user inside the product rather than regenerating from scratch: Vary (Region), released on 5 September 2023, lets a user select one area of an image and change only that; Style Reference lets a user upload an image and apply its style to a new one; Character Reference keeps the same character consistent across images; Image Weight controls how much influence an uploaded image has on the result.

Video arrived later and smaller than the image product. The V1 video model launched on 18 June 2025 — one week after Disney and Universal filed their copyright lawsuit — and it is image-to-video only: pick a still, press Animate, and get a 5-second clip, extendable in 4-second steps up to 21 seconds, at 480p (SD) or 720p (HD), with no sound and no text-to-video. A video job costs about eight times an image job, per the launch announcement. It is bundled into the same four subscription tiers, which is why the pricing page lists SD video on Basic and HD video from Standard up rather than selling video separately.

Pricing: four tiers, no free tier, and the meter is GPU time

There is no free tier to compare against. The pricing page offers four tiers, each with two prices: a monthly price and a 20% cheaper annual price. Basic is $10 a month or $8 a month billed annually, and covers about 200 image jobs a month — one job makes four images, so about 800 images — plus SD video generation, general commercial terms, optional Fast hours top-ups, 3 simultaneous Fast image jobs, 1 simultaneous Fast video job, and the Editor on uploaded images. Standard is $30 a month or $24 a month, and adds 15 Fast hours a month, HD video, unlimited Relax image generations and 3 simultaneous Fast video jobs. Pro is $60 a month or $48 a month, and adds 30 Fast hours, 12 simultaneous Fast image jobs, 6 simultaneous Fast video jobs, unlimited Relax image and SD video generations, and Stealth mode, which keeps generations private. Mega is $120 a month or $96 a month, and adds 60 Fast hours, 12 simultaneous Fast image jobs and 12 simultaneous Fast video jobs, with the same unlimited Relax generations and Stealth mode.

The annual commitment is a flat 20% on every tier, which works out to $96, $288, $576 and $1,152 a year by arithmetic on the printed prices — those annual totals are calculated, not printed on the page. The discount is the same at every tier, so there is no extra reward for buying the biggest plan, and the upgrade path is not price but capacity: hours, simultaneous jobs, video resolution and privacy. Basic is the only tier without unlimited Relax generations, which is the real wall between the $10 plan and the $30 plan — a Basic user who runs out of their 200 jobs waits or pays for top-ups, while a Standard user can keep generating slowly for free. Pro is the first tier with Stealth mode, which is the wall for professionals who do not want their generations visible to the community.

The comparison with the other AI image generation tools is where the model shows. Leonardo.ai publishes its prices — Essential $12 a month or $10 a month billed annually, Premium $30 or $24, Ultimate $60 or $48 — but only in a live browser: a plain text fetch of its pricing page shows the tiers and no price figures, which is why our master table recorded them as not publicly disclosed until 16 November 2026, when a live-browser read corrected the record. Krea.ai, NightCafe, Ideogram and Craiyon still do not publish prices on their pricing pages, and all four are recorded as not publicly disclosed in our master table. Midjourney and Leonardo.ai are the only two of the six whose price a visitor can actually read, and that transparency is itself a competitive fact: a buyer comparing the six can only complete the comparison on those two pages. Against the wider image tools, Basic at $10 a month is cheaper than Picsart Pro at $15 and PhotoRoom Pro at $12.99, but it covers image generation only, while those two bundle editing, video and exports. DeepAI Pro at $9.99 a month covers image, chat, video, music, voice and website building together for a dollar less than Midjourney Basic — the difference is that DeepAI sells breadth and Midjourney sells one model that looks better.

How does Midjourney get its users?

Not publicly disclosed. Midjourney publishes no acquisition mix, and our database holds no verified channel data for it, so this section is short on purpose. Anyone who tells you its traffic is exactly X percent organic search is guessing, however plausible the guess sounds.

Four things are observable without inventing numbers. First, Discord is the distribution. The product started as a Discord bot, and every generation happened in a public channel where other users could see it — the gallery was the marketing, and the community was the onboarding. The server launched on 14 March 2022, four months before open beta, which means the audience was built before the product was open to everyone. Second, the press did the rest. The Economist cover in June 2022, the Colorado State Fair win later that year, and the viral fake image of Pope Francis in a puffer jacket in 2023 each put the product in front of people who had never heard of a prompt. The last of those was not the company's doing, and it was not flattering, but it demonstrated the capability to millions of people at once. Third, the public surface is too thin for search to be the channel. The homepage as fetched is about twenty lines of About text, with no feature pages and no gallery, so the visits are not arriving through a catalogue of landing pages the way a PDF tool's visits do. Fourth, third-party reporting says the company spends nothing on marketing and has no marketing budget — that claim is third-party, not the company's own, and it is recorded with that label attached. If it is true, then every one of the 9.25M monthly visits arrived through the product, the community or the press, not through paid acquisition.

From visit to payment: the paywall is at the front door

The funnel is short because there is no free tier to graduate from. The pricing page lists no free tier and no trial length, so a visitor's first experience of the product costs $10 a month. Third-party reporting says free access was removed in March 2023 and the service has been subscription-only since, which matches what the pricing page shows today. What the page does not state, in the text fetched on 16 November 2026, is whether a card is required or whether any trial exists at all, so this teardown does not claim either way.

The paywall position is the front door, and the pricing is built to push the visitor toward the annual commitment at that point. Basic at $8 a month on an annual commitment against $10 a month billed monthly is a 20% discount for committing to a year, and the same 20% applies to every tier. That is a much smaller annual discount than the PDF tools offer — pdfFiller's annual commitment saves 60% — which suggests Midjourney does not need to bribe people into a year. The upgrade path is capacity, not price: a user who hits the 200-job cap, who wants unlimited slow generations, or who wants their work private has a reason to move to Standard at $30 or Pro at $60. What we cannot see from outside is the trial-to-paid rate — there is no trial to measure — the churn, or how much of the 9.25M monthly visits is paying subscribers versus people checking the price, so this teardown states the shape of the funnel and stops there rather than estimating a rate for it.

Retention: what keeps people coming back?

Not publicly disclosed. Midjourney publishes no churn, renewal or cohort figures in the sources we checked, so this section states what the product does to earn a return visit and marks the numbers as unknown.

Three features of the product support return visits anyway. The first is the version cadence. A new model version every few months — V7 in April 2025, V8 in March 2026, V8.2 in July 2026 — is a release event that brings lapsed users back to see what changed, and each version is visibly better than the last in a way a user can see in one generation. The second is the community. Generations happen in Discord channels and in the web gallery, where other people's prompts and results are visible, which turns a solitary tool into a place — and a place is harder to leave than a tool. The third is the subscription itself. A monthly payment is a monthly appointment, and the meter — Fast hours and job caps — is a reason to use the allowance before it resets. The counterweight is the same as for every image generator: generation is a commodity, and Leonardo.ai, Krea.ai, Ideogram, DALL-E, Stable Diffusion and Flux all do the same core action. A user who only generates plain images can switch in an afternoon, and the moat is taste — the look of the model — which is real but hard to measure from outside.

Why Midjourney is still standing

Our read, stated as a position you can argue with: Midjourney survives because it was profitable before it was big, and it never gave that independence away. The service entered open beta on 12 July 2022, and its founder told The Register the following month that the company was already profitable — within weeks of opening, not years. Its own About page calls it a lab of 60 people and a lean, self-funded, distributed team, and it has never taken venture capital. That means no board, no burn rate to defend and no pressure to grow at a loss. Third-party trackers estimate revenue in the hundreds of millions a year — roughly $300M to $500M for 2025, depending on the tracker — but they disagree with each other and the company publishes no official figure, so this teardown records the range and does not print a single revenue number. The counterweight to the independence is that the company is small: 60 people by its own account, against competitors with thousands of employees and billions in funding, and third-party trackers put headcount higher, at roughly 100 to 163. The two figures conflict, and this teardown uses the company's own.

The second half of the argument is the shape of the traffic. An estimated 9.25M visits in a single month makes Midjourney the fifty-seventh largest site in our database and the largest AI image generation tool, and the country split backs the story: the United States first at 28.61%, which is what an English-prompt product looks like, with China second at 6.62% and India third at 4.24%. The United States alone sends 2.65M visits, more than the whole traffic of CombinePDF at 2.5M, and India alone sends 392.17K visits, more than the whole traffic of Coggle at 362.99K. The caveat sits next to the number, as it does on every page of this site: the figure comes from a single third-party source for the month quoted, and Midjourney publishes no official MAU or DAU. October fell 16.78% from 11.11M in September, and no public source states a reason — a drop that size, unexplained, is a fact to watch rather than a trend to declare.

The last part of the argument is the model. A new version every few months, a video model added in June 2025, and a reputation for the best-looking images in the category are a product story that does not need a sales team. The risk in that shape is the same as the strength. A product whose only moat is taste can lose it in one release from a rival, and a company that has never taken outside money has no cushion if the lawsuits over training data go against it — Disney and Universal sued in June 2025, Warner Bros. Discovery in September 2025, and the cases were consolidated in November 2025. The defence is the community, the version cadence and the balance sheet, which is what independence buys.

What you can copy, and what you cannot

Copyable this week

Not copyable

Risks and gaps in this teardown

Similar sites in the database

Midjourney (9.25M) is the largest AI image generation tool in the database. Leonardo.ai (8.73M) is second, only 0.52M behind — 5.6% smaller. Krea.ai (5.5M) is third, about 1.7 times smaller, NightCafe (3.35M) fourth, about 2.8 times smaller, and Ideogram (3.1M) fifth, about 3.0 times smaller. Leonardo.ai now has a full teardown; the other three do not yet. DeepAI (18.17M) is the only AI generation tool larger than Midjourney, but it is a general AI platform, not an image generator, so the comparison is by pattern, not by category: Midjourney is the largest image-only generator in the database.

SiteCategoryMonthly visitsSource
Leonardo.aiImage tools8.73MSemrush, Oct 2026
Krea.aiImage tools5.5MSemrush, Oct 2026
NightCafeImage tools3.35MSemrush, Oct 2026
IdeogramImage tools3.1MSemrush, Oct 2026

Compare Midjourney with Leonardo.ai and Ideogram side by side. The full database, with working filters, is on the teardowns index.

Read the next one in your inbox

This teardown is free to read. The new ones go out by email first, one a week, with the traffic source and month attached to every number. Leave your address and the next one comes to you.