Midjourney and AI art: trends for 2027

For 2027, I expect better control over image editing and motion, alongside closer attention to content rights and data. This is a forecast based on the developments below, not a list of announced Midjourney releases.

One of the more interesting changes in image generation is fairly ordinary: more attention is going into fixing what has already been made. A pleasing first result is welcome. The real work begins when you need to change part of it and keep the rest.

That is where I would start a conversation about 2027. Next come costs, access to tools and rights to images. Here are a few developments that help make sense of those questions, followed by my own expectations.

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

A crowd facing a sheet of light. An illustration for the 2027 forecast.

Working on an existing image

On August 27, 2026, Midjourney opened its V8.2 editing model for testing. It supports text instructions and up to four reference images.^[1] September's changelog includes further editor and interface fixes.^[2]

This is a useful direction when you need a particular result. If the frame already works, you do not want to start again because of one detail. You want to work on that detail. An available editing feature is also different from a vision of future 3D worlds: the latter is not a release schedule.

Billions that are hard to compare

In its June 2024 release, Grand View Research forecasts a $1.08 billion AI image generator market in 2030.^[3] SkyQuest gives $30.02 billion for 2033.^[4] These are examples of forecasts with different horizons, not a measurement of the market in 2026. The different dates alone are a reason not to divide one figure by the other and present the result as a measure of disagreement.

A useful comparison also needs the scope and methodology of both reports. Their headlines cannot do that work. Both forecast growth, but neither figure tells a designer how many commissions to expect or what a client will pay for an image.

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

Several directions at once, with no single clear path.

A service can end before your project does

OpenAI closed Sora's app and website on April 26, 2026, with its API scheduled to close on September 24.^[5] That is enough reason to take dependence on a platform seriously. Even a familiar brand cannot guarantee that a service will stay available for years.

For a creator, the immediate question is whether you have your files and the materials needed to continue working. An account's generation history is convenient, but a poor substitute for your own archive. That applies to services we are happy with, too.

Copyright disputes do not have release dates

Disney and Universal filed their copyright complaint against Midjourney in June 2025.^[6] I would not treat 2027 as a guaranteed date for a judgment, or one case as the answer to every question about AI and authorship.

The platforms' responses may matter to creators as much as the eventual rulings. Changes to service terms can be felt before a case ends. For client work, check the conditions for your intended use rather than relying on a general claim about a commercial licence.

Labelling needs a closer reading

The AI Act sets different obligations for tool providers and people publishing certain content, including deepfakes.^[7] It does not amount to an identical caption requirement for every generated illustration.

The type of material and its publishing context matter. A realistic scene presented as a record of an event raises different questions from a fantasy illustration. Forecasts about next year can easily lose that distinction.

Where do the client's files go?

Discussion of European AI can sound like a contest over companies' nationalities. Yet one brand may use several suppliers' technology. In its February 2025 announcement, Mistral described using Black Forest Labs models for image generation in Le Chat.^[8]

A company's headquarters does not tell you where it processes data, how long it keeps it or what the contract allows. Those are the details to check for confidential client materials. A locally run model may be an option, but it also comes with hardware requirements and licence terms to review.

What will the audience see?

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

An AI-generated scene. What we choose to show also shapes its meaning.

Industry discussions often stop at the number of images generated. The audience rarely knows how many attempts we rejected or which model we used. They see the finished work. If a series is inconsistent or the idea unclear, another impressive detail will not fix it.

For commercial work, I would also look at revision time and the ability to maintain a chosen direction. Producing the first proposal quickly is only part of the assignment.

What I expect in 2027

These are my expectations, not product announcements. I expect editing and consistency across images to matter more. Video adds control over movement and timing, which a good still frame cannot resolve on its own.

I also think we will ask more questions about exporting files, usage terms and data handling when choosing a tool. Sometimes the reason will be as simple as needing to deliver a project and return to it six months later.

I do not know which company will come out ahead. For someone making images, learning to judge and improve their work seems more useful than correctly guessing the next market leader.

For practical exercises, read Midjourney in 2027: what to practise now. To work on your projects together, write to hello@midjourneyart.pl.

Sources checked on September 20, 2026. Forecasts for 2027 are the author’s assessment.