AI in practice

Midjourney ends free image creation, but finally comes to the web

Matthias Bastian
Midjourney locks free image generation and comes to the web

AI FAKE IMAGE / Midjourney prompted by THE DECODER

Midjourney will no longer offer free image generation for test accounts until further notice. The good news is: The AI image service will be coming to your browser soon, including social features.

According to Midjourney founder David Holz, there has been massive abuse of the free image generation by test accounts. Holz identified a viral tutorial video from China as a possible source.

Combined with lower GPU availability, this made Midjourney temporarily unavailable to paying customers. The restriction for free accounts will remain in place until a future system update.

The restriction has nothing to do with the fake images generated by Midjourney v5 for the arrest of Donald Trump or the Pope in a puffer jacket. For such scenarios, Holz promises improved moderation systems and an ongoing "thoughtful" approach. Community and expert advice will be part of the process.

The terms of service state that Midjourney may not be used to create drama, as Holz puts it. This includes the Trump pictures mentioned above. Midjourney maintains a country-specific list of forbidden words: after the Trump pictures, the word "arrest" was added to this list. However, such bans can be circumvented relatively easily by rewording the image titles, e.g. "handcuffed".

Furthermore, Midjourney cannot expand the list of banned words at will without massively limiting the creativity of the community and thus the usefulness of its own product. The new moderation system should help to keep the list of forbidden words to a minimum.

Midjourney gets web interface

In addition to the new moderation system, Midjourney is currently working on a new web interface that allows you to generate images directly in your browser. Currently, the Midjourney interface is only available for Discord users. The new web system will be rolled out gradually over the next two months and will offer social features in addition to image creation.

For now, Midjourney won't focus on AI image editing to compete with features like DALL-E 2's or Stable Diffusion's "Inpainting", which allows users to easily edit already created images using text, e.g. by enlarging or selectively replacing elements.

Midjourney continues to focus on image quality and speed. The latest v5 update demonstrates this by making image creation twice as fast at the same quality and halving the cost.