Midjourney
Midjourney is an AI-powered image generation platform that creates visual art from text prompts. It's designed for artists, designers, and creative professionals who want to explore AI-generated imagery and digital art creation.
Score generated by AI agents based on publicly cited evidence and reviewed by the project maintainer. Not independently validated.
Score History
Timeline events are AI-curated from public reporting. Score trajectory is derived from documented events.
Midjourney launched as a Discord-only open beta with a free 25-image trial and V3 model. The platform was novel and growing rapidly, reaching one million Discord users within months. Opacity concerns were minimal at this stage since the entire AI image generation field was new, though the closed-source model and Discord-only distribution already created modest lock-in. No lawsuits had been filed and governance structures were not yet under scrutiny.
V5's photorealism enabled viral deepfake incidents including the Pope Francis puffer jacket and fake Trump arrest images, forcing Midjourney to kill its free trial and begin crude word-banning as content moderation. The first artist class-action lawsuit was filed in January 2023, and Holz's December 2022 admission of training on 'a hundred million' unconsented images established the governance and legal trajectory. The platform was growing rapidly toward $200M annual revenue while remaining closed-source with no API.
The leaked 16,000-artist training spreadsheet, V6's near-exact reproduction of copyrighted content, and the banning of critic Reid Southen revealed deep governance problems. Midjourney updated its Terms of Service to shift copyright liability onto users while simultaneously declining to join the industry election deepfake pact. The CCDH study found Midjourney generated misleading election images 65% of the time. Content moderation through opaque word bans expanded, and the platform remained Discord-only with no API, deepening lock-in for its growing commercial user base.
Disney, Universal, and Warner Bros. filed and consolidated major copyright suits describing Midjourney as 'a bottomless pit of plagiarism.' Stanford's transparency index scored Midjourney 14/100, the joint-lowest among all evaluated AI companies. EU AI Act compliance obligations took effect with up to 7% revenue penalties. V7's mandatory personalization profiling deepened opacity, while the web app launch partially offset Discord lock-in. Revenue reached $500M on ~130 employees with no VC funding, but the legal and regulatory walls continued closing in.
Alternatives
Free tier with no minimum subscription — addresses Midjourney's removal of its free trial head-on. Particularly strong at text rendering in images, which Midjourney has historically struggled with. Committed to responsible AI training practices and clearer commercial licensing terms. Easy switch via web interface; quality is competitive though the style differs.
Adobe's AI image generator trained exclusively on licensed and public domain content, which sidesteps the copyright litigation that has Midjourney facing consolidated suits from Disney, Universal, and Warner Bros. Has a free monthly credit allowance with no subscription required to start. Quality is strong for commercial use cases; outputs come with Content Credentials documenting AI generation. Easy switch for casual users, though the aesthetic differs from Midjourney's style.
Dimensional Breakdown
Summaries below were written by AI agents based on the cited evidence. They are editorial interpretations, not independent research findings.
Dimension History
Timeline (31 events)
Midjourney launches closed beta with no model documentation
Midjourney launched its closed beta Discord server, requesting users post high-quality photographs for systems training. From its earliest release, the platform operated as a fully closed-source model with no published documentation about architecture, training data sources, or model parameters. Unlike Stable Diffusion, which would later release its model weights and architecture details, Midjourney disclosed nothing about how its AI generated images.
Midjourney enters open beta on Discord
Midjourney launched its open beta as a Discord-only AI image generation service with version 3, offering 25 free image generations to new users. The platform was accessible exclusively through Discord bot commands, requiring users to join the Midjourney server. Within months, the Discord server surpassed one million users.
AI-generated Midjourney art wins Colorado State Fair
Jason Allen won first place in the digital arts category at the Colorado State Fair with 'Theatre D'opera Spatial,' an image generated primarily using Midjourney. The judges were unaware the entry was AI-generated. The $300 prize sparked global controversy about AI's role in art, with artists furious over the precedent. The U.S. Copyright Office later denied copyright protection for the work.
Midjourney V4 launches with photorealistic capabilities
Version 4 launched in alpha, bringing significantly improved photorealism and image coherence. The model became the default on December 20, 2022. V4's increased realism made Midjourney outputs harder to distinguish from real photographs, raising the stakes for deepfake and misinformation concerns while also making the closed-source model more valuable and harder to replicate.
Holz admits to training on 'hundred million' images without consent
PetaPixel reported on a Forbes interview in which Midjourney founder David Holz acknowledged the company used 'a hundred million' images scraped from the internet without creator consent to train its AI models. When asked about obtaining permission from living artists, Holz stated there 'isn't a way to get a hundred million images and know where they're coming from,' and confirmed creators cannot opt out of their work being included.
Artists file first copyright class-action lawsuit
Three artists -- Sarah Andersen, Kelly McKernan, and Karla Ortiz -- filed a class-action copyright infringement lawsuit against Stability AI, Midjourney, and DeviantArt in the U.S. District Court for the Northern District of California. The suit alleged the companies trained AI image generators on copyrighted works scraped via the LAION-5B dataset without artist consent, and that the resulting tools could produce work in artists' distinctive styles.
Midjourney bans reproductive anatomy terms
MIT Technology Review revealed that Midjourney had banned a wide range of words relating to the human reproductive system, including 'placenta,' 'fallopian tubes,' 'mammary glands,' 'uterine,' and 'cervix.' The bans were identified as disproportionately targeting female anatomy. Founder David Holz described the bans as a 'stopgap measure' while the company improved AI-side moderation, but the approach drew criticism for being crude and overreaching.
V5 launches with near-photorealistic output quality
Midjourney released version 5, which pushed image generation toward high photorealism, more accurate prompt interpretation, and broader stylistic flexibility. V5 became the default model on March 30, 2023. The quality leap made Midjourney outputs increasingly difficult to distinguish from real photographs, directly enabling the deepfake incidents that followed within days.
Viral AI-generated Pope Francis puffer jacket image
A Midjourney V5-generated image of Pope Francis wearing a white Balenciaga puffer jacket went viral across social media, amassing over 20 million views and fooling millions into believing it was real. The creator, a Chicago construction worker using the handle Pablo Xavier, was subsequently banned from Reddit. The incident was described as 'the first real mass-level AI misinformation case' and drew global attention to AI deepfake risks.
Midjourney kills free trial citing abuse and demand
Midjourney permanently ended its free trial program, which had previously offered 25 free image generations to new users. Founder David Holz cited 'extraordinary demand and trial abuse' as the reason, noting that the viral deepfake images of Pope Francis and Donald Trump -- generated days earlier -- had driven a surge of new sign-ups and that many users were creating multiple accounts to exploit free credits. All users now require a minimum $10/month Basic subscription.
Washington Post reports Midjourney 'making fake images mainstream'
The Washington Post published an investigation into how Midjourney's AI image generator had become the primary tool driving viral deepfake imagery into mainstream media. The report detailed how the platform's photorealistic V5 model produced convincing fake images of Trump's arrest, the Pope, and other public figures, and examined Midjourney's inadequate content moderation response to the crisis.
Fake Pentagon explosion image causes stock market dip
An AI-generated image purporting to show an explosion near the Pentagon went viral on social media, including from a blue-check account impersonating Bloomberg News. The S&P 500 briefly dipped before officials confirmed no incident had occurred. While not definitively attributed to Midjourney, the incident demonstrated the real-world consequences of realistic AI image generation and increased regulatory scrutiny of platforms like Midjourney.
Midjourney remains Discord-only with no API as competitors launch integrations
By mid-2023, OpenAI had launched its DALL-E API and Stability AI had made Stable Diffusion open-source, but Midjourney continued to operate exclusively through Discord bot commands with no official API. The Terms of Service explicitly prohibited automation, and businesses caught using unofficial third-party API wrappers faced immediate account bans. This Discord-only lock-in forced users to build workflows around a platform they could not programmatically access, making switching increasingly costly as prompt libraries and generation histories accumulated.
Judge dismisses most claims against Midjourney in artist lawsuit
U.S. District Judge William Orrick dismissed all allegations against Midjourney and DeviantArt in the Andersen v. Stability AI class-action suit, granting leave to amend the complaint. The court found the original claims lacked specificity regarding Midjourney's use of copyrighted works. However, the copyright infringement claim against Stability AI for training Stable Diffusion survived, keeping the broader legal theory alive.
Expanded artist class action adds 4,700 artists and Runway AI
The artists' legal team filed an amended complaint in the Andersen v. Stability AI case, expanding the class to include over 4,700 artists and adding Runway AI as a new defendant alongside Midjourney, Stability AI, and DeviantArt. New plaintiffs included prominent artists Grzegorz Rutkowski, Gerald Brom, and Jingna Zhang. The expanded complaint included 455 pages of supplementary evidence documenting how artist styles were allegedly curated for training.
V6 alpha reveals near-exact reproduction of copyrighted content
Midjourney released V6 in alpha, and concept artist Reid Southen quickly demonstrated it could produce near-identical reproductions of copyrighted film stills when prompted with character names and movie titles. A prompt for 'Joaquin Phoenix Joker movie, 2019' generated an almost identical still from the 2019 film. Midjourney banned Southen within 12 hours, wiped his prompting history, and canceled his subscription. The company simultaneously updated its Terms of Service to shift copyright liability onto users.
Leaked spreadsheet reveals 16,000 artists in training data
A Google spreadsheet titled 'Midjourney Style List' surfaced on social media, containing over 16,000 artist names allegedly curated for training Midjourney's models. The list included prominent living and deceased artists such as Banksy, David Hockney, Yayoi Kusama, and a six-year-old hospital fundraiser. Screenshots showed a 2022 message from founder David Holz welcoming the addition of 16,000 artists. The Art Newspaper and The Register confirmed the list also appeared in court filings for the class-action lawsuit.
Midjourney refuses to join election deepfake pact
At the Munich Security Conference, 20 major tech companies including Adobe, Amazon, Google, Meta, Microsoft, OpenAI, and TikTok signed a voluntary pact to combat AI-generated deepfakes targeting elections. Midjourney was the only maker of a leading AI image-generating tool that declined to participate. CEO David Holz indicated that election-related content 'wasn't the purpose of Midjourney' and that he didn't prioritize political speech concerns.
CCDH study finds Midjourney generates election deepfakes 65% of the time
The Center for Countering Digital Hate published its 'Fake Image Factories' report finding that Midjourney produced misleading election images in 65% of test cases -- the highest rate among all AI image tools tested (compared to 38% for Image Creator, 35% for DreamStudio, 28% for ChatGPT). Researchers demonstrated that Midjourney's ban on Biden and Trump images could be trivially bypassed by adding a single backslash to prompts or using physical descriptions instead of names.
Midjourney blocks Biden and Trump images before election
Midjourney CEO David Holz announced the platform would block users from generating images of President Biden and former President Trump ahead of the 2024 U.S. presidential election. Holz described it as a temporary measure, acknowledging that 'this moderation stuff is kind of hard.' The action came a month after Midjourney refused to sign the industry deepfake pact and following the CCDH report showing high deepfake generation rates.
Judge rules artists may pursue copyright claims against Midjourney
U.S. District Judge William Orrick ruled that artists including Sarah Andersen, Kelly McKernan, and Karla Ortiz could proceed with their copyright infringement claims against Midjourney, Stability AI, DeviantArt, and Runway AI. The judge found the artists had 'reasonably argued' that AI companies violated their rights by storing copyrighted works and that Stable Diffusion 'may have been built to a significant extent on copyrighted works.' Trial was set for September 8, 2026.
Midjourney launches standalone web app beyond Discord
Midjourney released its web-based interface alongside version 6.1, marking the platform's first move beyond Discord-only access. The web app consolidated editing tools including panning, zooming, region variation, and inpainting into a single interface. Access was initially limited to users with over 10,000 generated images, later reduced to 1,000. While improving accessibility, the web migration reportedly broke access to legacy archive endpoints used by third-party export tools.
Reports mount of bans without explanation, refunds denied
By late 2024, user complaints about Midjourney's opaque ban and refund practices had escalated significantly. Multiple accounts on Medium, Trustpilot, and BBB documented users being banned immediately after paying for annual subscriptions, losing $300+ with no explanation. Users who relied on Fast Mode too frequently reported being accused of unauthorized automation despite not violating Terms of Service. The appeals process directed users to a bot-run system or outdated Google form with up to two-week response times, and appeals were rarely acknowledged or overturned. The refund policy -- limited to accounts that have used fewer than 20 GPU minutes total -- ensured most banned users had no recourse.
Midjourney reaches $500M revenue on 130 employees with 60% margins
Midjourney's revenue reached $500 million in 2025, a 66.7% increase from $300 million in 2024, achieved with approximately 130 employees and no external venture capital. Revenue per employee exceeded $5 million annually, and projected profit margins reached 60% with 95% gross margins on inference. While the bootstrapped model avoided VC extraction pressure, the company's pricing -- $10 to $120 per month with no free tier since April 2023 -- extracted substantial revenue relative to its lean cost structure, particularly from users with no viable alternative that matched Midjourney's aesthetic quality.
V7 launches with mandatory personalization profiling
Midjourney released V7, the first version to have personalization enabled by default. All users must rate approximately 200 image pairs before using the model, building an aesthetic profile that Midjourney uses to tailor outputs. The profiling process is opaque -- users cannot inspect how their preferences are interpreted or stored. Early testers reported 'poor, aesthetically subdued results,' and reviewers noted that quality improvements had plateaued relative to competitors like FLUX and DALL-E 3.
Disney and Universal sue Midjourney for copyright infringement
Disney Enterprises and Universal City Studios filed a copyright infringement lawsuit against Midjourney in the U.S. District Court for the Central District of California, describing the AI company as 'a bottomless pit of plagiarism.' The suit alleged Midjourney unlawfully trained its AI on copyrighted works and produced infringing outputs featuring Disney and Universal characters. This marked the first time major entertainment studios directly sued an AI image generator.
Midjourney launches video generation model V1
Midjourney released its first video generation model, expanding from still images to five-second image-to-video clips extendable to 21 seconds. Video jobs cost approximately 8x more than image jobs. The expansion into video generation increased the platform's competitive scope but also expanded the surface area for copyright and deepfake concerns, as the same training data and content moderation challenges now applied to moving images.
EU AI Act obligations take effect for GPAI models
The EU AI Act's transparency and compliance obligations for general-purpose AI models took effect, requiring providers like Midjourney to maintain technical documentation, publish training data summaries, prepare transparency reports, and appoint EU representatives. Non-compliance carries penalties of up to 35 million euros or 7% of global revenue. Midjourney, which scored 14/100 on Stanford's transparency index, faces among the largest compliance gaps of any evaluated AI company.
Warner Bros. Discovery sues Midjourney for 'systematic' infringement
Warner Bros. Discovery filed a separate copyright infringement lawsuit against Midjourney in the Central District of California, alleging the AI service was 'developed using illegal copies of Warner Bros. Discovery's copyrighted works.' The complaint included examples of AI-generated content featuring characters including Superman, Batman, Wonder Woman, Bugs Bunny, Scooby-Doo, Tom and Jerry, and the Powerpuff Girls, claiming Midjourney enables subscribers to generate infringing images and videos of these characters.
Judge consolidates Disney and Warner Bros. suits against Midjourney
Judge John A. Kronstadt consolidated the Disney and Warner Bros. copyright lawsuits against Midjourney for all purposes including trial, with the Disney case designated as the lead case. The court ordered the parties to private mediation by August 19, 2026, with a post-mediation status conference set for August 31, 2026. The consolidation signaled the growing legal weight of entertainment industry opposition to AI image generation trained on copyrighted content.
Stanford transparency index rates Midjourney lowest at 14/100
Stanford's 2025 Foundation Model Transparency Index scored Midjourney at 14 out of 100, tying with xAI for the lowest score among all 20 evaluated AI companies. The index found Midjourney disclosed no information about its training data, model architecture, model size, risk assessments, or environmental impact. As Midjourney did not submit a voluntary transparency report, Stanford researchers had to manually assess the company based on publicly available information.
Evidence (40 citations)
D1: User Value Erosion
D2: Business Customer Exploitation
D3: Shareholder Extraction
D4: Lock-in & Switching Costs
D5: Twiddling & Algorithmic Opacity
D6: Dark Patterns
D7: Advertising & Monetization Pressure
D8: Competitive Conduct
D9: Labor & Governance
D10: Regulatory & Legal Posture
Scoring Log (4 entries)
Added 2 missing dimension narratives