Midjourney opens its public beta

Background

A text-to-image model learns from large collections of pictures paired with written descriptions, and once trained it can turn a short piece of text — a prompt — into a new picture. For a company building one, the hard part is not only the model but giving people a place to use it. Midjourney's answer was to skip a standalone website or app: the whole service ran inside Discord, an existing chat platform where people already gathered in shared channels.

The beta opens

On 12 July 2022 Midjourney opened its image generator to the public as an open beta ^[1][2]. People used it through a bot on Midjourney's Discord server ^[2]: a user typed a prompt, and the bot posted the resulting images back into the conversation.

What using it was like

Because the front end was the Discord chat service, generating an image was a social act rather than a private one. New users submitted their prompts in newbie channels, and they saw their results alongside other users doing the same thing ^[4]. Anyone in the channel could watch a prompt being tried, see what came back, and adjust their own wording in response — a very different experience from typing into a form on a website.

There was a way out of the crowd. Paying subscribers could send their prompts to the Midjourney bot in a private Discord direct message instead of a public channel ^[4]. Privacy, in other words, was one of the things a subscription bought, along with the image generation itself.

Versions and price

The software did not stand still during the beta. V3 was released on 25 July 2022, two weeks into the open beta ^[2]. Access was sold by subscription at $10 or $30 a month ^[3], so the service was paid rather than free from the start, with the lower tier as the entry point.

Scale

The open beta brought users quickly. By August 2022, Holz said the service had hundreds of thousands of users and needed about 10,000 servers ^[3]. That number is a reminder of what image generation costs to run: each picture requested by a user has to be computed on hardware that the company pays for, and demand from hundreds of thousands of people translates into a large and expensive fleet of machines.

Why it mattered

Three things about this launch shaped how the service was experienced. First, it was open: an open beta meant the generator was available to the general public rather than to a hand-picked group ^[1][2]. Second, the interface was borrowed rather than built — a bot inside a chat service — which made prompting visible and collective, and made a private direct message a paid convenience ^[4]. Third, the business model was a subscription from the beginning, at two price points ^[3], while the compute behind it was heavy enough to require thousands of servers within weeks ^[3].

For people who arrived during those weeks, the entry point to AI image generation was not a product page but a chat channel full of strangers' prompts and the pictures they produced.