Midjourney 5 Has Arrived - Here is What Has Changed
Just a few new features but overall improved images.
Midjourney has introduced an alpha version of Midjourney 5, and the image quality is noticeably better. This is surprising to say, given that version 3 was very good and version 4 was a notable upgrade. According to Midjourney’s release announcement in its Discord server:
What's new with the V5 base model?
Much wider stylistic range and more responsive to prompting
Much higher image quality (2x resolution increase) improved dynamic range
More detailed images. Details more likely to be correct. Less unwanted text.
Improved performance with image prompting
Supports
--tile
argument for seamless tiling (experimental)Supports
--ar
aspect ratios greater than 2:1 (experimental)Supports
--iw
for weighing image prompts versus text prompts
The higher image quality is also accompanied by the generated images that are upscaled by default. For Midjourney 4 and earlier versions, lower-resolution images were initially generated, and then you could request an upscaled image. However, it is the versatility and detail that has veteran users excited.
What’s Better?
Olivio Sarikas, a popular YouTuber focused on generative AI text-to-image models, mentioned in his review the clear improvements in hands, teeth, and photorealism. And photorealism is really where it is easiest to see the improvements. The people in the images on this page are not real.
Another often-cited difference is the level of detail. Midjourney 5 can produce greater detail and greater coherence in the images. This generally means there will be fewer errors and incoherent elements. Some of the detail in earlier versions looked fine until you tried to inspect the image more closely.
What are Some Key Differences?
Aside from the subjective differences that are only apparent when viewing the art, there are some other notable changes in version 5.
More supported styles - Midjourney says this version is less opinionated on what image should be produced. This means it is more likely to follow your prompts faithfully, including the requested style. The downside is that longer prompts may be required to achieve similar results in version 5. It also claims to have more dynamic range which also can help support additional styles.
Higher default resolution - Midjourney 4 images were 512 x 512 pixels by default. In version 5, they are twice the resolution at 1024 x 1024 pixels. You needed to upscale a selected image to get that resolution in version 4.
More support aspect ratios - Midjourney 4 launched with one aspect ratio and later expanded to five. Midjourney 5 appears to support whatever ration you choose, though this is considered “experimental” by the company.
V4 is Still Very Good
So, from a subjective standpoint version 5 is better, but it is also true that version 4 is very good. Midjourney 4 still compares favorably to the new version, thought you can see how the level of detail sharpness and dynamic range have evolved with the new model.
At this point, text-to-image generators are becoming so good that quality is not the key area of development required to advance further. The biggest areas for improvement continue to be coherence and control. Midjourney 5 made notable advances in both areas while also improving many aspects of image quality.
Let me know what you find in your own use of Midjourney 5 and how you think it compares to Stable Diffusion and DALL-E.