Every year around this time the internet floods with "trends for next year" pieces - mostly hunches dressed up as a list. This one is meant to be different: every sentence below either has a source or is clearly flagged as my own synthesis. The subject: where AI image generation, and Midjourney specifically, is actually headed - based on what already happened in 2026, not on what someone imagines.

Method: only dated, sourced facts - model versions, rulings, laws, financial figures. Wherever I move into prediction, I say so directly.

A crowd of silhouettes standing in shadow before a bright red glowing surface, every figure turned to face the light

Three independent sources, three different forecasts for the same market - and yet all of them facing the same direction

Where Midjourney actually stands

The default model is V8.2, live since July 24, 2026 - focused on aesthetics and personalization, described by the company as "more creative, bold, sophisticated, edgy"[1]. The Edit Model arrived August 27, replacing the previous Omni Reference and Character Reference tools - editing an existing image from written instructions, up to four reference images at once, built into the editor[2]. The most recent changelog (September 16) is minor: Korean language support, an "Enhance" button replacing the old Upscale[3] - no new model version since July.

More interesting is the direction CEO David Holz states outright: images, then video, then 3D, ultimately "real-time open-world simulations." His words: "I think the 3D model and the faster model (real-time) will probably come sooner [than people expect]"[4]. The company is also building a hardware team for unspecified projects tied to spatial computing[5].

The market: nobody agrees on the numbers

This is exactly the point where most "trends" articles paste one number from one report and fake certainty. Reality: ask three analyst firms how big the "AI image generator market" is, you get three completely different numbers.

  • Grand View Research: $349.6 million (2023) → $1.08 billion by 2030[6]
  • Precedence Research: $3.16 billion (2025) → $30.02 billion by 2033[7]

That's the same market segment, roughly a nine-fold gap. The mundane reason: each firm defines what counts as the "image generator market" differently. One number worth trusting more, because it comes from a broader, more established methodology: Bloomberg Intelligence puts the entire generative AI market at $2.3 trillion by 2032, revising its own year-old forecast upward by half a trillion[8]. Practical takeaway: read every "the AI market will be worth $X" headline alongside who calculated it and how they defined "market" - the number alone, without that context, means nothing.

A chaotic, painterly composition with a large colourful eye and the handwritten digits 2027 in the corner, full of dense, tangled lines and scribbles

This is roughly what trying to forecast this market's size looks like - plenty of lines, none of them pointing at the same place

The Sora warning: what it costs to be a video-generation god

The year's most telling story: Sora, OpenAI's video model, launched publicly February 1, 2026. Shutdown was announced March 24. The app and web version stopped working April 26 - after 84 days. API access ends September 24, 2026[9] - essentially as you're reading this. The reported reason: unsustainable compute costs amid chip shortages[10].

Video generation is technically possible today, but economically brutal. The models that survived are winning in different ways: Kling AI (Kuaishou) leads on revenue - $153 million in fiscal 2025, roughly a $500 million annualized run-rate by April 2026[11]. Runway leads on valuation - $5.3 billion after a Series E round in February 2026[12]. Google's Veo 3.1 leads on distribution, partly through integration with "Nano Banana" - Google's own image model, which generates keyframes that Veo then animates[13].

The courts will decide more than the models

The case most directly relevant to Midjourney: Disney, Universal and Warner Bros. Discovery sued the company in 2025 for copyright infringement (characters like Superman and Batman showing up in generated images). Discovery closes September 2026 - now. Summary judgment motions are due November 23, 2026. A real jury trial, if the case doesn't resolve earlier, is more likely in 2027[14] - which is itself a forecast worth noting: this dispute will probably define 2027 for the whole industry before it's even settled.

Context from neighboring cases: Anthropic's $1.5 billion settlement (given final approval July 20, 2026) is the largest publicly known copyright settlement in US history[15] - a real signal of what training on illegally-sourced data can cost. In the NYT v. OpenAI case, the US Department of Justice sided with OpenAI in September 2026, arguing national security interest[16] - a sign the US government doesn't necessarily want these cases to end in penalties for AI companies.

Regulation: the world is teaching AI to say where it came from

Article 50 of the EU AI Act (the AI content labeling obligation) took effect August 2, 2026 and - unlike the rules for high-risk systems, which got pushed to December 2027 - this deadline was NOT delayed[17]. The related Code of Practice has already been signed by roughly 190 organizations, including Google, Meta, Microsoft, OpenAI and Anthropic[18].

Midjourney isn't on that list - and, more pointedly, the company hasn't implemented C2PA (image-provenance metadata), despite being a Content Authenticity Initiative member since 2023. That directly exposes it to California's SB 942, active since August 2, 2026, with penalties up to $5,000 per day per violation[19]. For comparison: OpenAI has embedded C2PA in DALL-E 3 images since February 2024.

China has its own, more elaborate version of the same idea since September 1, 2025 - mandatory explicit (visible) and implicit (metadata) labels for all AI-generated content[20]. The direction for 2027 is clear regardless of jurisdiction: labeling stops being a best practice and becomes a legal obligation - and companies that ignore it start paying for it.

Sovereignty: who's actually answering this question

A separate thread that deserves its own section: pressure is building for AI data and compute to stay in Europe, under European control, not on American servers. This isn't a vague declaration - on June 3, 2026 the European Commission unveiled a tech sovereignty package (Chips Act 2.0, the Cloud and AI Development Act, an open-source strategy)[25]. Earlier, under the InvestAI initiative, €200 billion is on the table - €50B public, €150B private[26]. On July 30, 2026 a call opened for seven "AI gigafactories": the EU is putting up to €5B on the table, member states match it, expected private backing is over €20B - together, more than €30 billion into infrastructure alone[27].

The face of this sovereignty push today is Mistral AI. On September 8, 2026, literally days before this piece was published, the company raised €3 billion in a round led by Samsung, at a valuation above €21 billion - the largest funding round in the history of a European tech company[28]. France's Ministry of the Armed Forces signed a framework agreement with Mistral covering every branch of the military, hosted on France-controlled infrastructure, explicitly to avoid dependency on the US[29].

There's a catch most "European AI sovereignty" pieces skip: Mistral has no image-generation model of its own. Le Chat's image feature runs on FLUX, a model built by Black Forest Labs - which happens to be a genuinely European company, headquartered in Freiburg, Germany, that raised $300 million in December 2025 at a $3.25 billion valuation[30]. If you're looking for Europe's answer to Midjourney, it isn't Mistral - it's Black Forest Labs.

So where does that leave Midjourney itself? Nowhere, for now. OpenAI turned on an EU data-residency option back in February 2025, explicitly covering image content too[31]. Midjourney, based on the analyses available, doesn't offer that option - data sits on US servers, with no standard data-processing agreement for businesses[32]. It's the same pattern as the C2PA gap described earlier: competitors are investing in European compliance, Midjourney isn't, at least not yet.

A separate, smaller thread is local processing - images generated directly on a phone or laptop, with nothing sent to any cloud. Apple built this into iOS 27, partly hybrid since some tasks still route to a server[33]. Apps like Draw Things have run Stable Diffusion or FLUX fully offline on an iPhone for years[34]. This is a separate trend from European sovereignty - the motive here is more privacy and cost than politics - but the direction is the same: less dependence on someone else's cloud.

Stepping past the facts here: I couldn't find a single source directly linking EU regulation to rising demand for local image processing specifically - these are two separate, real trends, not one proven causal chain. If Midjourney keeps ignoring both sovereignty and content labeling, though, the natural consequence is that European companies and public institutions needing EU-resident data will pick OpenAI or Black Forest Labs instead - not because they're technically better, but because they answer a question Midjourney is currently dodging.

The creative industry: from experiment to infrastructure

Adobe Firefly: 75% of Fortune 500 companies use it, 24+ billion assets generated, AI-linked revenue has crossed $5 billion - over a third of Adobe's total revenue[21].

A surreal, vividly coloured scene: a woman in a kimono and a dark, horned creature sit side by side on a red couch beneath a green eye painted on the wall, a cat between them

This is roughly what the creativity in this section looks like today - one prompt, zero genre boundaries

Cannes Lions 2026 introduced a new "AI Craft" award category for work that "couldn't exist without" AI[22]. WPP consolidated production into a new "WPP Production" unit, with an annual generative-AI budget around £250 million[23].

But it's not purely a success story. Coca-Cola's holiday ad (2025), entirely generated by three AI studios using four different models, saw positive sentiment drop from 23.8% pre-launch to 10.2% after[24]. AI in advertising has moved from experiment to infrastructure - that doesn't mean audiences have fully accepted it yet.

What this means for 2027 - three predictions

From here I'm moving past pure fact - this is my synthesis, not a citation:

  1. The battlefield shifts from static images to video and real-time. Holz's stated direction, Sora's death as a cost warning, the Kling/Runway/Veo race - these aren't separate stories, they're one story about where the money and attention are going.
  2. AI content labeling stops being an EU curiosity. California and China already have their own versions. At the rate other states and countries are copying these rules, 2027 is the year skipping a label starts costing real money, not just inside the EU.
  3. The Disney/Universal/WBD v. Midjourney ruling sets a precedent for the whole industry - regardless of who wins. It's the first major case where the defendant makes images, not text, and the court schedule points to a 2027 resolution.
Want to learn Midjourney before the market accelerates even further? Check out szkoleniemidjourney.pl or write: hello@midjourneyart.pl

Frequently asked questions

Will Midjourney really build "real-time worlds"?

That's the CEO's stated vision, not an announced product with a release date. Treat it as a direction, not a promise.

Does Sora's death mean the end of AI video?

No - it means the end of ONE model that cost too much relative to revenue. Kling, Runway and Veo keep growing, each on a different financial path.

Can these market forecasts be trusted?

The gap between analyst firms (up to nine-fold for the same segment) is a reason to treat any single number with caution, not a reason to dismiss the trend itself - the direction of growth is consistent even where the numbers aren't.

Sources

  1. Wikipedia, "Midjourney"
  2. Powerdrill, "Midjourney V8.2 Edit Model: Presentation Visuals, Access, and Alternatives"
  3. Midjourney, "Alpha Changelog - 9/16/26"
  4. Level Up Coding, "David Holz Shared His Long-Term MidJourney Vision"
  5. Press Farm, "Midjourney's Next Act"
  6. Grand View Research, "AI Image Generator Market Size To Reach $1.08Bn By 2030"
  7. SkyQuest, "AI Image Generator Market"
  8. Bloomberg Intelligence, "Generative AI Market Poised to Reach $2.3 Trillion by 2032"
  9. Wikipedia, "Sora (text-to-video model)"; OpenAI, "What to know about the Sora discontinuation"
  10. TechCrunch, "OpenAI's Sora was the creepiest app on your phone - now it's shutting down"
  11. Sacra, "Kling revenue, funding & news"
  12. Morphed, "Runway Statistics 2026"
  13. Google AI, "Nano Banana image generation"
  14. The Art Newspaper, "Midjourney strikes back: sued AI giant demands Hollywood's secrets"
  15. Open Class Actions, "Anthropic $1.5B Copyright Settlement Wins Final Approval"
  16. Washington Post, "DOJ urges judge to rule for OpenAI, Microsoft in N.Y. Times lawsuit"
  17. Gibson Dunn, "EU AI Act Omnibus Agreement - Postponed High-Risk Deadlines"
  18. European Commission, "Strong backing for the Code of Practice on Transparency of AI-generated Content"
  19. Neomanex, "CA AI Transparency Act Takes Effect - Midjourney Lacks C2PA"
  20. China Law Translate, "Measures for Labeling of AI-Generated Synthetic Content"
  21. Futurum Group, "Adobe Q4 FY 2025: Record Revenue, AI Adoption, ARR Targets"
  22. Haute Living, "Cannes Lions 2026 Unveils Creative Brand Lion Award"
  23. Marketing Dive, "WPP rolls production capabilities into new WPP Production unit"
  24. Content Grip, "Coca-Cola's AI holiday ad triggers backlash"