首页AI 新闻
[v5 Related] Midjourney v5 vs. v4 Generated Image Comparison

[v5 Related] Midjourney v5 vs. v4 Generated Image Comparison

Setsu1
Setsu1
2023-03-31 16:45

Midjourney v5 is finally here, and this article will share with you the latest updates and enhancements of this version compared to the previous version v4. Fasten your seatbelts and let's explore the new changes of Midjourney v5 together!

More realistic output images

In addition to generating more realistic and detailed images, the new model also requires more precise instructions. Midjourney stated that they used new neural architecture and new aesthetic techniques to develop this model, which was trained for about five months on Google Cloud's "AI supercluster." This model can produce more realistic images and more accurate details. In the first test, with no clear image style prompts, v4 produced an illustration, while v5 generated a more realistic museum installation for the prompt "made from banknotes." Thanks to v5, the results are also closer to the expected outcome.

Midjourney v4:

Midjourney v5:

Clearer and more detailed images

Since upgrading to Midjourney, you will notice that your images are now clearer and more detailed. This is the benefit brought by the new image processing algorithm implemented in this latest version. As you can see when comparing V4 and V5, V5 creates clearer and more detailed images. Here is a comparison of V4 and V5 for the prompt "two children playing with Lego bricks." At first glance, V4's image looks realistic.

Midjourney v4:

Midjourney v5:

However, upon closer observation, we can see that the image looks more like cartoon characters, with the children's faces looking like rubber masks and their hands being a mess while the Lego bricks look like blurry copies. V5, on the other hand, looks more like a real photo, with amazing level of detail.

Wider range of supported style

The Midjourney team claims that V5 has five times the style range of previous generations, which means you can use longer, more descriptive prompts to create various visual effects, from landscapes to buildings, and more. The algorithm has also been optimized for natural language processing for more accurate results. In addition, many issues of V4 have been fixed, such as better handling of a large number of people, more realistic hands, more accurate finger counts, and less image defects.

Using image prompts to generate more "accurate" and predictable results

Compared to previous generations, Midjourney V5 offers a wider variety of images, and is no longer limited to specific art styles. Holz suggests that longer prompts are required to describe details in the image, such as lighting and ambiance, to achieve the desired effect. Holz also stated that the current V5 model is an alpha version, and will undergo major changes before the final release. Like previous versions, the final version will have default styles that are more suitable for beginners. Creatives should not expect to see the current V5 style in the future.

Biomorphic Skyscraper and Snow Mountain:

Higher resolution of MidJourney V5

MidJourney users have been hoping for upgraded image resolution, and we are pleased to announce that the team is gradually achieving this goal. By default, MidJourney V5 images have a resolution of 1024 x 1024 pixels, while V4 has 512 x 512 pixels. V4 images must be upgraded to this resolution, so this is a significant improvement. In the future, we may see the upscaler of V5 capable of raising the resolution to 2048 x 2048 pixels. However, for now, 1024 x 1024 pixels is the default and maximum size.

全部讨论

no data来都来了,坐下聊聊