Names, versions and owners change fast. Checked October 2026.
In plain words
Midjourney is a text-to-image tool. You type a description, such as "a lighthouse at dawn, watercolour", and a minute later it shows you four different pictures to choose from. You can then enlarge a favourite or ask for variations on it.
The name covers both the company and the service. Midjourney, Inc. is a small independent lab in San Francisco, founded by David Holz. It trains its own image models, which are not downloadable. The service opened to the public in July 2022, first through the chat app Discord and later on its own website.
Its models are numbered versions, and a new one arrives every few months; the box above shows the newest. Since June 2025 it can also turn a picture into a short video clip. Under the hood it is a kind of generative AI for images.
Everyday example
A sketch artist at a mela
At a fair, you describe a scene to a quick sketch artist. They dash off four rough versions. You point at the one you like: "this one, but with the moon bigger." They redraw it larger and in more detail. Midjourney follows the same loop: describe, see four drafts, pick, refine.
Where the example breaks: The sketch artist understands your story and remembers how real moons look. Midjourney matches patterns from images it was trained on, so it can still get hands, text or counts of objects wrong.
Latestas of October 2026
- Newest image model
- Version 8.2, released 24 July 2026
- Video
- Image-to-video model, available since 18 June 2025
- Company
- Midjourney, Inc., independent, led by founder David Holz
These details change often. The rest of this page explains what it is, which stays true.
How it actually works
Midjourney does not publish full details of its models, but the general method is known.
Your words become numbers
The prompt is turned into an embedding: a list of numbers that captures what you asked for, which guides the picture.
A picture grows out of noise
Image models of this kind usually start from random noise and remove it step by step, steering each step toward the prompt. This is how a diffusion model works.
Four drafts, then refinement
Each request returns a 2 by 2 grid. Choosing one lets you upscale it to higher resolution or make four new variations close to it. Settings let you control style, shape and how far it strays.
Learned from many images
The models learned from very large collections of images with descriptions. Several film studios, including Disney and Universal in June 2025, have sued Midjourney over the use of their characters. These cases are part of a wider debate on AI and copyright.
- Made by
- Midjourney, Inc., San Francisco, founded by David Holz
- Open to the public
- 12 July 2022, as an open beta on Discord; website from August 2024
- Models
- Its own image models, released as numbered versions and not downloadable
- Video
- Image-to-video model, first released 18 June 2025
Questions people ask
Why has Midjourney been sued?
Film studios say its models were trained on, and can reproduce, their copyrighted characters. Disney and Universal sued in June 2025 and Warner Bros. Discovery in September 2025. These cases are part of a wider legal debate on AI and copyright.
Is Midjourney the same as DALL-E or Stable Diffusion?
No. All three turn text into images, but they come from different makers. DALL-E was OpenAI's image model, now replaced by GPT Image. Stable Diffusion is from Stability AI and can be downloaded. Midjourney's models stay on its own servers.
Do you still need Discord to use Midjourney?
No. Midjourney started inside Discord, where you typed a command to a bot. Since August 2024 it also has its own website where you can type prompts, browse and edit your images.
Who owns Midjourney?
Midjourney was founded by David Holz as an independent company in San Francisco, not as part of a big technology group. Ownership and leadership can change, so the box above shows its current status.