AI tools

What is Midjourney, and how do you get started?

· By Robert Scales · 4 min read

Short answer

Midjourney is an AI image generator from an independent research lab in San Francisco, and on July 12, 2022 it opened its beta to the public. You use it inside the chat app Discord by typing /imagine followed by a description, and it returns four images in about a minute.

Two days ago Midjourney opened its beta to everyone. Until now you needed an invitation; today anyone with a Discord account can join its server and start making pictures from text. Midjourney is the name of both the tool and the small company behind it, a San Francisco research lab led by David Holz, who previously co-founded the hand-tracking company Leap Motion.

The tool has been drawing attention during its invite-only phase. In June, The Economist used a Midjourney image for one of its covers. The company has been updating its model every few months since the first version appeared in February.

Where Midjourney lives

Unlike most software, Midjourney has no app of its own for creating images. Everything happens in Discord, a free chat platform organized into servers and channels. You type commands to the Midjourney Bot, a program that reads your message and posts the results back in the channel. A website at midjourney.com shows a gallery of your own images and other people's.

According to Midjourney's quick start guide, new members can make a limited number of images, about 25, before they need a paid subscription. Images made in public channels are visible to everyone there, which is a good way to learn from others but means you should not type anything private.

Your first image, step by step

The quick start guide lays out a simple path:

  1. Join the Midjourney Discord server with your Discord account.
  2. Open a newbies channel, such as #newbies-1. Bot commands only work in bot channels, not in general ones like #support.
  3. Type /imagine, then press space, Tab or Enter to open the prompt field, and write your description.
  4. Wait about a minute. The bot replies with a grid of four images.
  5. Upscale or vary. Buttons U1 to U4 make a larger version of one image, around 1024 by 1024 pixels, with added detail. Buttons V1 to V4 make new variations of one image.
  6. Save it. Open the full-size image and save it to your computer, or react with the envelope emoji to have the bot send it to you in a direct message.

A first prompt might look like this:

Prompt
/imagine prompt: a lighthouse on a rocky island during a storm, crashing waves, moody oil painting

Midjourney's code of conduct is short and worth reading before you start. Its first rule reads "Don't be a jerk," and it bans gore, adult content and images meant to upset or inflame.

Try it: Write one subject, such as "a cozy library," and run it three times with a different style at the end each time: "watercolor," "photograph," "pixel art." Compare the grids and note which words changed the image most.

Tips for better results

Midjourney rewards the same habits as any image generator. Name the subject first, then the setting, then the style and mood. Use concrete words for light and color ("golden hour," "neon," "muted pastels"). When a grid has one promising image, use its V button rather than starting over; variations let you explore close to something that already works. Keep a note of prompts you like, since small wording changes can shift the result a lot.

How does Midjourney compare with DALL·E 2?

Both turn text into images, but they differ in access, workflow and focus. OpenAI announced DALL·E 2 in April and is still offering it as a research preview to a limited group of users, with a waitlist for others. Midjourney, as of this week, lets anyone join and try a few free images right away.

MidjourneyDALL·E 2
AccessOpen beta on Discord, free trial then subscriptionLimited preview, waitlist
How you use it/imagine commands in chat channelsInvited testers only; not offered in the API
Main toolsFour-image grids, upscale and variation buttonsRealistic images, editing parts of a photo with words, variations
RulesCode of conduct bans gore and adult contentContent policy bans violent, adult and political images; blocks realistic faces of real people

The workflow feels different too. Midjourney is social: you create in public channels, watch other people's prompts appear, and learn by example. DALL·E 2's preview focuses on what OpenAI calls editing, such as adding or removing an object in an existing photo while keeping the lighting consistent, an ability Midjourney's guide does not describe.

Which is better depends on your goal. If you want to start experimenting today and like learning alongside other people, Midjourney is the one you can actually use now. If you need photographic realism or edits to your own pictures, DALL·E 2 is built for that, once you get in. Many people will likely end up using both, much as illustrators keep more than one brush.

Sources

  1. Quick Start Guide - Midjourney Documentation (July 2022) — Midjourney, July 14, 2022
  2. Midjourney — Wikipedia, July 12, 2022
  3. DALL·E 2 — OpenAI, April 6, 2022

Get the free Guide to Artificial Intelligence

Sign up for the weekly newsletter and receive the free guide: what changed in AI this week, one lesson, one prompt to try. Unsubscribe in one click.

How we handle your address

Back to top

Display

Your choices are saved in this browser only. No data is sent.

Text size
100%
Colours
Reading
More settings
Accessibility statement