Midjourney 5.1 is designed to deliver better results with simple prompts. The developers have also added a new moderation system.
Midjourney users can now access and use Midjourney 5.1, which can be enabled in the Discord settings. According to Midjourney, v5.1 is much easier to use than 5.0 and should deliver high-quality images even with short prompts.
However, the images are more “opinionated”, similar to v4, the company writes, which could mean that creative freedom is somewhat limited. Midjourney overcomes this with a “RAW mode” in the settings that makes 5.1 more “unopinionated”, comparable to the standard mode from 5.0 that is still available. The RAW mode is available in the preferences after activating v5.1.
Otherwise, v5.1 is said to offer greater consistency, better prompt accuracy, fewer unwanted borders or text artifacts, and increased sharpness compared to v5.0.
Midjourney bans word ban
Midjourney came under fire in March when fake photos created with MJ v5 began circulating on social media, including one of Donald Trump’s alleged arrest.
Midjourney more or less successfully prevented the creation of more fake photos by blocking words and generally maintaining a list of words for which the system refuses to generate images or even issues warnings to the point of banning.
But this sledgehammer moderation limits creative freedom: the word “arrest,” for example, which was blocked after the Trump incident, can be used in many contexts, including legitimate ones.
Midjourney wants to solve this problem with a new “AI moderation”: it can understand words in context and is, according to the company, “much more intelligent and nuanced” than just blocking words.
The AI moderator works in two steps: First, it performs a quick scan of a prompt to check for violations. If there is an alert, the user can object. The objection is then reviewed by a more powerful AI system.
Because the second step is “quite expensive,” Midjourney limits the number of objections to 25 per day. Successful objections do not count against this limit. The company plans to adjust the limit based on traffic and cost trends. The AI moderator effectively replaces the word blocking system for v5 and older models, effective immediately.
“We’re hoping that this two layer system gives people an almost 100% chance of getting innocent prompts through,” the announcement reads.
Midjourney v5.1 and the new moderation system are currently being tested. If all goes as planned, both will be rolled out as the new default by the end of this week.