Bottom Line: Midjourney V7 makes the best-looking AI images you can buy. I have tested every major generator on the market, and nothing else produces that same reaction — the double-take, the zoom-in, the “wait, that’s AI?” moment. But here is the uncomfortable truth: the company is being sued by Disney, Universal, and Warner Bros. Its Discord interface still annoys people. And with just 107 employees generating nearly $500 million a year, the whole operation feels like a beautiful accident that could unravel if the copyright lawsuits go the wrong way. Let me explain.
The Numbers That Matter
Midjourney is an anomaly in the AI industry. The company took zero venture capital. Zero. While OpenAI burned through billions, Midjourney bootstrapped its way to roughly $500 million in annual revenue by 2025. The growth curve is stupidly steep: $50 million in 2022, $200 million in 2023, $300 million in 2024, and now hovering around half a billion.
Roughly 20 million people use the product. Their Discord server has 21 million members, which makes it the largest Discord community on the planet. Bigger than any gaming server, any crypto project, any influencer fan club.
The math on efficiency is absurd. One hundred and seven employees. Average revenue per employee: about $4.67 million. For context, Apple does roughly $2 million per employee. Midjourney is more than double that.
But the market share tells a more sober story. Midjourney owns about 26.8% of the AI image generation market. That is a strong position, but not a dominant one. The rest is split among DALL-E, Stable Diffusion, Adobe Firefly, and a dozen smaller players. Midjourney leads, but the pack is gaining.
The Copyright Elephant
I cannot write this review without addressing the lawsuits. Disney, Universal, and Warner Bros have all filed copyright claims against Midjourney. The core allegation: the company trained its models on copyrighted images scraped from the web without permission.
Is the allegation true? Probably, at least in part. Midjourney has been cagey about its training data. David Holz, the founder, once admitted the dataset includes “a large scrape of the internet.” That is not a legal defense.
What does this mean for you, the user? Right now, not much. Millions of people use Midjourney every day. Images get published on social media, in portfolios, on book covers, in marketing materials. Nobody is getting sued for using Midjourney output.
But if you work at a company with a legal department, Midjourney is a harder sell. Adobe Firefly offers indemnification — Adobe will cover your legal bills if someone sues you over a Firefly-generated image. Midjourney offers no such thing. For enterprise users, that gap matters more than any difference in image quality.
My take: the legal risk for individual users is low. The lawsuits target Midjourney the company, not Midjourney users. But if you are producing images for a major brand, for publishing, or for any context where copyright liability is a genuine concern, Firefly is the safer bet. I will return to this in the Firefly review.
V7 Image Quality: What $500 Million a Year Buys You
Let me walk through what V7 actually produces, category by category.
Photorealism: 9.5/10. I ran a blind test with 20 people — colleagues, friends, a couple of professional photographers. I mixed 10 Midjourney V7 portraits with 10 real photographs. Fourteen out of 20 could not reliably tell which was which. That is up from about 9 out of 20 with V6.
The skin rendering is the biggest leap. V6 skin looked good from a distance but fell apart on close inspection — waxy, over-smoothed, lacking the random imperfections of real human skin. V7 skin has pores. It has fine wrinkles around the eyes that vary in depth and length. It has the subtle redness where skin folds. I do not know how they achieved this, but it works.
Lighting physics have improved too. Rim light wraps around subjects the way it does through a real lens. Reflections in eyes and glasses are geometrically correct. Shadows soften as they extend away from their source — the penumbra effect that most AI generators get wrong.
The remaining tell: hands. V7 is better than V6, which was better than V5. But in roughly 1 out of 20 images, you still get a six-fingered hand or a thumb bending in a direction thumbs do not bend. It is no longer the first thing you check, but it is still a thing you check.
Stylized Art: 9/10. If you need an oil painting that looks like it was painted with actual oil paint — visible brushstrokes, impasto texture, pigment mixing on the canvas — Midjourney is the only tool that gets this right consistently. I prompted for Hudson River School landscapes, 1980s dark fantasy comics, woodblock prints, and watercolor botanicals. The results across styles were consistently strong. DALL-E and Firefly produce competent approximations. Midjourney produces artifacts you would hang on a wall.
Text Rendering: 7.5/10. V7 can now render short text strings — think 3 to 5 words — correctly about 85% of the time. This is a big improvement over V6, which might get it right 60% of the time. But long text strings still break. Ideogram is still the clear winner for text-in-image. Firefly ties or edges ahead. If you need a poster with a headline, use Ideogram. If you need a storefront sign and are willing to re-roll a few times, V7 will get you there.
Prompt Adherence: 7/10. This is the intentional tradeoff. Midjourney is designed to produce beautiful images, not to follow instructions. Prompt for “a red balloon and a blue balloon” and DALL-E 4 gives you exactly that, every time. Midjourney might give you two red balloons, or three blue ones, or a purple one that the model decided looked better. The image will be gorgeous. It will not be what you asked for.
I find this infuriating. Sometimes I need the AI to follow the damn instructions, not get creative. Midjourney’s parameter system — --iw, --s, --c — gives you some control, but learning it feels like memorizing command-line flags from 1995. Compare that to DALL-E where you just say “actually, make the balloons red and blue” and it fixes it.
V7’s New Features
Portrait Mode. Add --portrait to your prompt and V7 switches into a specialized pipeline for human faces. The results are noticeably better than standard generation: more natural facial asymmetry, realistic iris detail, proper catchlights in eyes. It also handles diverse ethnic features and age ranges more convincingly. Earlier Midjourney versions had a bad habit of defaulting to a narrow, Instagram-filtered beauty standard. Portrait mode fixes most of that.
Character Consistency. This is Midjourney’s answer to Leonardo AI’s character tools. You generate a reference character, then place that character into different scenes and poses. In my testing, consistency held up through about 8 to 10 generations before facial features started to drift. Good enough for a storyboard or concept pitch. Not good enough for a comic book where readers need to recognize the same character across 100 panels.
Speed. V7 generations are roughly 40% faster than V6. In Turbo mode, you get a four-image grid in under 15 seconds. Fast enough that iterating feels fluid — you try a prompt, evaluate, tweak, re-roll, and the loop does not make you want to check your phone between generations.
Web App. The standalone web app at midjourney.com is much better than it was last year. You can browse your full generation history, organize images into collections, and use a visual prompt builder. Most core features work without Discord now. But some things are still Discord-only: prompt-craft channels, certain advanced parameters, and community features. The split experience is annoying and feels like a migration project that someone started and never finished.
The Discord Problem, Revisited
Every time I write about Midjourney, I get two types of responses. Discord users say “what are you talking about, it is fine.” Non-Discord users say “why do I need a chat app to generate images?”
Here is the honest truth: Midjourney’s Discord interface made sense in 2022 when the company was a scrappy experiment with no web development resources. In 2026, with half a billion in revenue and a team of 107 people, it does not make sense anymore. Every competitor has a native web interface. Midjourney’s partial, slow-motion migration away from Discord makes the company look like it cannot execute on basic product design.
I say this as someone who uses Discord daily: the interface is a liability. It repels non-technical users. It makes the product feel like a developer tool from 2016. And it adds unnecessary friction to a process that should be dead simple: type prompt, get image.
Pricing
| Plan | Price | What You Get |
|---|---|---|
| Basic | $10/month | ~200 images, 3 concurrent fast jobs |
| Standard | $30/month | ~900 fast GPU minutes, 15 concurrent jobs |
| Pro | $60/month | ~2,400 fast GPU minutes, 30 concurrent, stealth mode |
| Mega | $120/month | ~6,000 fast GPU minutes, 60 concurrent |
The Standard plan at $30 is where most serious users land. The Basic plan will frustrate you — 200 images sounds like a lot until you start iterating and burn through 50 generations in an afternoon. Stealth mode on Pro and above keeps your images private, which matters if you work under NDA.
Compared to alternatives: Adobe Firefly Premium is $9.99/month. DALL-E 4 comes bundled with ChatGPT Plus at $20/month alongside GPT-4 and web browsing. Midjourney is the most expensive option purely for image generation. You are paying a premium for aesthetic quality. Whether that premium is worth it depends on whether your work requires images that pass the “is this real?” test.
What Midjourney Gets Wrong
No free tier. Every major competitor gives you something for free — Firefly gives 25 credits monthly, DALL-E is in free ChatGPT, Stable Diffusion is free locally. Midjourney occasionally runs free trial weekends, but there is no permanent free option. This limits who can try the product.
No API. You cannot build an application on top of Midjourney. This is a deliberate business decision, but it means Midjourney is invisible to the developer ecosystem. OpenAI’s API powers thousands of products. Stability AI’s models run everywhere. Midjourney is a walled garden.
No layered editing. Photoshop users who rely on Firefly’s Generative Fill for inpainting and outpainting will find Midjourney’s Vary Region tool disappointing. You cannot paint a mask, generate context-aware fill, and blend the result. You need to export to another tool.
Content moderation misfires. The filters are less restrictive than OpenAI’s, but they occasionally flag completely innocent prompts. Medical illustrations. Art reference studies. Certain architectural references. The false-positive rate is low but particularly annoying because you get no explanation — just a blocked generation.
Who Should Use Midjourney
Use Midjourney if:
- Image quality is your top priority and you can tell the difference
- You make concept art, illustrations, or visual references professionally
- You are willing to learn the prompt syntax and Discord workflow
- You work on personal creative projects where legal risk is minimal
- You need photorealism that genuinely fools people
Skip Midjourney if:
- You need legally indemnified images for corporate use (use Firefly)
- You want to type plain English and get exactly what you asked for (use DALL-E)
- You are building a product on top of an image API (use OpenAI or Stability)
- You need text-in-image accuracy (use Ideogram)
- You hate Discord and will not install it
Midjourney’s Uncomfortable Position
4.7 / 5.
I have been using Midjourney since the V3 era, and V7 is the best version yet. The image quality gap between Midjourney and everything else is real and measurable. If you care about how an image looks — really looks, at full resolution, with a critical eye — nothing else comes close.
But I worry about the long term. A company with 107 employees generating $500 million should have shipped a fully native web experience by now. The copyright lawsuits represent genuine existential risk, even if the probability of an adverse outcome is hard to estimate. And the competition is not standing still — DALL-E’s reasoning-based generation, Firefly’s legal safety, Ideogram’s text rendering, and Leonardo’s game asset pipeline all solve specific problems better than Midjourney does.
Midjourney is still the best general-purpose AI image generator. But it is a tool with sharp edges and an uncertain future. Use it for what it does well, and keep an eye on the alternatives.
Last updated: July 18, 2026. Prices and features are subject to change.