Midjourney New Features – What’s new?
Table of Contents
Midjourney AI is an ever-evolving tool, constantly under pressure to retain its title within the zeitgeist as the best AI art generator for photorealism, and perhaps overall. As such, their developers are working fast to implement Midjourney New Features that keep them one step ahead – and we keep this article up to date every time they do.
What Midjourney New Features came out this month?
Midjourney is set to announce new upscaling algorithms this weekend, along with a mysterious upgrade called “Personalization”. It’s unclear at this time what the latter could do to your AI-image generation experience.
- Subtle Upscaler – “a more faithful upscale, reduces pixelation but keeps any glitches in images.”
- Creative Upscaler – “a more dramatic change, more likely to change details and if it sees an issue it will remove it.”
No maximum resolution has officialy been announced, though it’s expected to be 2x the previous maximum.
The new personalization feature is quietly being rolled out to select users, with a wider release expected later this month.
What Midjourney New Features came out in 2023?
Pan
Users can now “pan” their generated images. With the new directional buttons provided after upscaling an image, you can use Midjourney to expand the image in one direction, revealing what it imagines would exist outside of the original framing. This alters the aspect ratio of the original image, but the final output resolution is limited to the same number of total pixels as it was prior to panning. This prevents infinite image resolutions running server bandwidth into the ground, and running the company into bankruptcy.
Essential AI Tools
Zoom
Users can also zoom-out existing upscales. Sadly, this doesn’t apply to images generated prior to the update. This process is being culturally adopted under the term “outpainting”, coined originally by OpenAI for their DALL-E back in 2022, and Adobe also offer their own implementation via Firefly called “Extend image“. Also colloquially known as “uncropping”, the new zoom out feature is exactly the same as the pan functionality but applied in every direction at once.
Selecting the “Zoom Out 2x” or “Zoom Out 1.5” options will maintain the aspect ratio of your upscale, whereas “Custom Zoom” allows you to specify both the degree of zoom (– – zoom 2 being 200%) and aspect ratio. One use case for this functionality is to make square images out of accidentally rectangular generations, without having to crop in.
Video capabilities
Under the radar, Midjourney has added – –video functionality to versions 4 and 5. This implementation isn’t usable to the same degree as their static image offering, but is a promising indication of ongoing development in generative video AI.
Midjourney v5.2
The recent release of version 5.2, the new flagship model, broadly means that everything looks better than previous versions. Specifically though, Midjourney cites sharper images “with better colors, contrast, and compositions” as well as output images responding more noticeably to the – –stylize parameter, essentially a high variation mode.
Midjourney v5.1
Prior to that, version 5.1 is generally agrees to have made a significant leap in photo realism. This is proving to be one of the last great hurdles in generative imaging AI, alongside typeface-crisp text in a real human language, and human hands with no more or less than five fingers. Much to the dismay of creatives (Twitter users) who’s reputation relies on them being able to tell the difference, real photos and their AI-generated counterparts are no longer consistently distinguishable to the human eye.
From this point on, Midjourney has responded more empathetically and with greater understanding to simple or vague text prompts, where previously it’s interpretation could be compared to a vindictive genie. It now produces results of “high Coherency, excels at accurately interpreting natural language prompts, produces fewer unwanted artifacts and borders” and allows repeating patterns with – –tile.
Remix mode
Without remix mode off (default) clicking the “Make Variations” button under an upscaled image will provide 4 variants of that image. The framing, art style, and subject matter will be identical but you’ll notice differences in the details. With remix mode on, however, clicking that same button will produce a pop-up window which provides the user a chance to alter the text prompt, which can lead to an entirely different image with none of the similarities mentioned above.
The main benefit of this is to avoid typing out the same prompt by hand to make minor tweaks, which only becomes a notable relief when dealing with long prompts or several repetitive uses of near identical prompts. To use remix mode, type “/prefer remix” within Discord. You can then type the same again to toggle it off.
About the Author