Midjourney: images with a strong style
What Midjourney is good at, how its prompting differs from a chat model, and the licensing questions that matter.
Pricing
Paid plans from about US$10/month; no free tier (check the official page)
Official site
Open the official website (opens in a new tab)
Prices and plans change. Check the official page before you pay.
Midjourney generates images and has a recognisable aesthetic: rich lighting, confident composition, and a tendency towards the cinematic. It is the tool many designers reach for when they need a mood rather than an exact illustration.
What it is good at
- Concept art and mood boards. A look and feel to discuss, before anyone commits to a direction.
- Editorial and abstract imagery. Textures, backgrounds and atmospheric scenes that would take hours to illustrate.
- Style exploration. The same subject in five styles, quickly, to see which one communicates.
- Iteration. Each result suggests variations, which makes exploring cheap.
Where it fails
- Text in images. It has improved and still misspells words, especially long ones.
- Precise composition. "A red cup on the left, a blue book on the right" is a suggestion, not an instruction.
- Hands, jewellery and fine mechanical detail, which still produce artefacts at small sizes.
- Accuracy. It illustrates a plausible version of a subject, not a correct one; a historical scene will contain invented details.
Prompting
Midjourney prompts are a list of descriptors rather than sentences: subject, setting, lighting, medium, style, and parameters. Aspect ratio and stylisation are set with flags such as --ar 16:9 and --stylize.
Start short and add. A six-word prompt with a clear subject often beats a paragraph, and the parameters do more work than adjectives.
Pricing
There is no free tier. Plans start around US$10 per month for a limited number of image generations, with higher tiers for more fast hours and commercial terms. Check the official page for the current allowances, which are measured in GPU minutes rather than image counts.
Licensing
Paid subscribers own the images they create, subject to the terms, with broader rights on higher tiers; companies above a revenue threshold need a higher plan. The terms have changed before, so read them before a commercial launch.
There is also a copyright question separate from the licence: in several jurisdictions, an image generated entirely by a model may not be protectable. For a brand asset, that matters.
Who it is for
Designers, illustrators and marketers exploring a direction. It is not a tool for a diagram with labels, a product photo that must match reality, or anything where a wrong detail is a legal problem.
Alternatives
DALL·E inside ChatGPT is better at following a literal instruction and at rendering short text. Adobe Firefly is designed for commercial safety with training data Adobe can license, and it integrates with Photoshop. Stable Diffusion gives full control if you are willing to run it yourself.