AIWiki
Malaysia

Midjourney

An independent AI research lab and image generation service that produces images and video from natural-language text prompts, accessible primarily through Discord and a web application.

5 min readLast updated May 2026Companies & Tools

Midjourney is an independent AI research lab and the name of the generative image service it produces. Founded in 2021 by David Holz, a co-founder of Leap Motion, Midjourney is privately held, self-funded, and has remained one of the most influential generative image platforms despite its small headcount and lack of venture capital backing. The service produces images and short video clips from natural-language text prompts and is best known for the painterly, stylised aesthetic of its default outputs.

Service and access

Midjourney is delivered primarily through a Discord bot interface, although a dedicated web application at midjourney.com and a mobile app are also available. Users issue prompts through slash commands, and the service returns a grid of four candidate images that the user can upscale, vary, or refine. Subscription tiers are billed monthly and govern queue priority, monthly generation hours, and access to relax mode.

Unlike services such as DALL-E or Stable Diffusion, Midjourney does not publish its model architecture, training data, or weights. The service is closed-source and proprietary, with model improvements rolled out through versioned releases.

Model versions

Midjourney has shipped sequential numbered model versions since 2022. V4 introduced significantly better coherence; V5 brought photorealism; V6, released in late 2023, added accurate in-image text rendering and stronger prompt adherence. Midjourney V7 entered alpha in April 2025 and became the default model in June 2025, adding improved realism, better handling of hands and bodies, a Draft Mode that generates ten times faster at half cost, and Omni Reference for character consistency. The V1 Video Model arrived in June 2025, allowing users to animate static images into short clips. V8 entered alpha in March 2026, and V8.1 added native 2K high-resolution generation without upscaling.

| Version | Release | Notable additions | |---------|---------|-------------------| | V5 | March 2023 | Photorealism | | V6 | December 2023 | In-image text, prompt adherence | | V7 | April 2025 (alpha) | Draft Mode, Omni Reference | | Video V1 | June 2025 | Image-to-video animation | | V8 / V8.1 | March 2026 | Native 2K resolution |

Aesthetics and prompting

Midjourney has long been associated with a distinctive house style — rich colour, cinematic lighting, and a slight painterly quality — although recent versions render across a wider stylistic range. Prompting is controlled through natural-language descriptions augmented by parameter flags that control aspect ratio, stylisation strength, chaos, and weight of reference images. Character Reference and Style Reference features allow users to anchor outputs to a supplied image while varying composition.

Midjourney has been a defendant in several copyright lawsuits brought by visual artists, stock-image companies, and film studios who allege that the service was trained on copyrighted works without authorisation. The legal landscape for generative image services remains unsettled in the United States, the European Union, and across Asia. The company has not publicly disclosed its training corpus.

See Also

References

References

  1. Midjourney. (2024–2026). Official Documentation and Version Release Notes. docs.midjourney.com.
  2. Holz, D. (2022–2024). Office Hours Transcripts. Midjourney Community Discord.
  3. Andersen, S., et al. v. Stability AI, Midjourney, DeviantArt. (2023). United States District Court, Northern District of California.
  4. Malaysian Communications and Multimedia Commission. (2024). Discussion Paper on Generative AI and Online Content. MCMC.
  5. Intellectual Property Corporation of Malaysia. (2023). Annual Report on Copyright and Emerging Technologies. MyIPO.