Midjourney opens its beta on Discord

Midjourney, a small independent lab led by David Holz, opened its image-generation service to open beta, run entirely through a bot on the chat platform Discord. Users typed an /imagine command followed by a text prompt; the bot returned four candidate images, which could then be upscaled or used to generate variations. Unlike OpenAI’s approach with DALL·E 2, launched three months earlier to a small waitlisted cohort with heavy filtering, Midjourney’s default behaviour posted both prompts and results into shared public channels.

That default made image generation visible and social rather than private. Newcomers to a Discord server could watch a stream of other users’ prompts and outputs in real time, learn what phrasing produced striking results, and iterate on each other’s ideas — a discovery mechanism no closed, single-user interface offered. Midjourney also developed a distinctive house aesthetic, generally favouring dramatic, painterly, high-contrast compositions, which made its output recognisable and gave it an identity distinct from more photorealistic competitors.

The Discord-native approach let a small team scale quickly with minimal separate infrastructure, and Holz told The Verge the following month that the company was already profitable, without having raised outside venture funding. Midjourney’s imagery began circulating widely beyond its own community within weeks, including winning a fine-arts competition prize under a human entrant’s name later that year, an episode that helped ignite a broader argument about AI-generated art, authorship and copyright that continued through subsequent releases of Stable Diffusion and DALL·E 2’s own wider rollout.