Midjourney releases V8.2 image model
Midjourney has released V8.2, a point update to its image model that targets aesthetics, fewer low-quality generations and better personalization, though the company gave no benchmarks or measured results.
Midjourney announced the launch of its V8.2 image model on Friday. The company described it as an update focused on "aesthetics and image quality and personalization."
According to the announcement, the release brings four main changes:
- Aesthetics: Midjourney says images should be "more creative, bold, sophisticated, edgy and fresh."
- Reliability: It says random cases of low-quality output "should be dramatically reduced."
- Personalization: The model should "understand" a user's taste much better, especially for users who have rated many images on their profiles.
- Profile creation: People building personalization profiles for V8.2 now choose from a "much larger and improved pool of images."
Midjourney credited user image ratings for making the update possible. It encouraged users to try both new and existing personalization profiles with the new version.
The announcement is short and gives no quantitative evidence. It includes no benchmarks, failure-rate figures or before-and-after comparisons. It also doesn't say how much less often low-quality images appear, what changed in training or architecture, or whether pricing, speed or default model settings are different. The improvements are Midjourney's own description and have not been independently checked.