On December 21, Midjourney announced the release of the latest beta version, v6, on Discord. This model, trained over 9 months, lacks clear benchmark comparisons but, based on current performance demonstrated through online tests, can be considered the most advanced Vincentian graph model available. The quality of the samples in this release not only disrupts the design sphere but also directly challenges the photography industry.
On December 21, Midjourney announced the beta release of its latest version, v6, on Discord. This model, trained over 9 months, lacks clear benchmark comparisons, but based on its current performance demonstrated through online tests, it can be considered the most advanced Vincentian graph model available. The quality of the samples in this release not only disrupts the design sphere but also directly challenges the photography industry.
Portrait Through the Evolution of AI:
According to the official introduction on Discord, the improvements in Midjourney’s model this time are mainly in the following aspects:
- Greater Accuracy and Support for Longer Prompts
Based on user tests who participated in the beta, Midjourney v6 now supports prompts up to 350 characters long; the current version can specify more topics/colors and other details in the prompts. The improvement in following instructions accurately is also evident from online tests. For instance, when tested in French, the images generated by the v6 model matched the specific nuance of the prompt much more closely: featuring boxing gloves and a slight smile.
While v5.2 struggled with understanding actions like lying down and then wanting to stand up and roll back, v6 handles these scenarios seamlessly.
2) Improved Coherence and Model Knowledge
According to the announcement, Midjourney states:
- v6 now responds more quickly to prompts. Avoid using vague terms like “impressive, realistic, 4k, 8k.”
- Be clear about what you want. While it may not be as exciting, it will help you better convey your intentions.
User testing shows that v6 can even understand punctuation and grammatical nuances. This upgrade will significantly impact the overall user experience. Prompts that were previously like alchemy can now be replaced with more detailed descriptions. It’s hard to say whether this increases or decreases the threshold for newcomers.
In response to this change, a designer has shared new tips and techniques that are currently most suitable for Midjourney: the core is clear, non-technical, structured expression.
A small but notable detail in v6’s model knowledge development is illustrated by the following example: the prompt here is “Japanese woman.” v6 can distinguish subtle differences in appearance between various Asian ethnic groups. Compared to v5.2, the woman in the image has more distinctly Japanese features. The AI has leveraged differences in Asian appearances that Europeans might not fully comprehend.
3) Improved Prompting and Image Blending
The improvements in image prompting specifically refer to enhancements in generating detailed and realistic images based on text prompts. The enhanced image blending means that v6 can make combinations of different elements and styles more harmonious and natural.
According to user trials, the image generation capability of v6 has significantly improved. Compared to v5.2, images produced by v6 are indeed more rich, detailed, and vibrant. While v5.2 was already quite “real” in its detail, v6 gives a more “authentic” feel rather than just a “simulation.” With the correct prompt, the results can even have a real texture right from the camera.
For instance, between these two plates of mashed potatoes and beef stew, the v6 version on the right is clearly more appetizing
Regarding environmental restoration, the difference is also quite clear: the image below shows a Roman market with v6, which would make you think it’s a scene from a “Rome” film. However, v5.2 appears a bit flat, with somewhat odd-looking buildings.
The contrast between these two sets of images is even more pronounced. The prompts are documentary-style, and the top images truly look like documentary footage, whereas the lower ones still have many issues upon closer inspection, including a lack of realistic lighting, shadows, and depth of detail.
In terms of portraiture, v5 was already quite restored. However, in v6, the small imperfections on the face, wrinkles, and stains on the brim of the hat, along with the changes in lighting and shadows near the bridge of the nose, make the images from v5.2 look flatter in comparison. Even with freckles, v6’s performance remains more convincing, with full focus matching human body appearance rather than a flat feeling.
4) Improved Text Rendering
Simply put, v6 can now render text in images more clearly and even decide its style. For the best results, the text should be placed in quotation marks. This helps the model distinguish between descriptive signals and actual text presented in the image. For example, if a user wants to create an image with the text “Hello World!” written in marker on a sticky note, they can use a prompt like: “Imagine a picture with the text ‘Hello World!’ written in marker on a sticky note –ar 16:9 –v 6.” This has been clearly demonstrated in testing. The accuracy of text handling in v6 has greatly improved, and the embedding feels better.
In another example, the Coca-Cola font has been perfectly restored.
This is another logo design—the text fits perfectly with the style of the image. The designer’s final position is somewhat shaky.
The powerful capabilities that Midjourney v6 demonstrates seem to pose a real threat to photographers and design teams—its ability to create harmonious and aesthetic atmospheres is now on par with high-end photographers. What’s most frightening is not just the power it shows, but the speed of its evolution. In Midjourney’s announcement, they also noted that the speed, image quality, coherence, timely updates, and text accuracy of v6 will improve in the coming weeks. The beta v6 release had its first update half an hour after launch, and creation speed has increased by 2.7 times.
With the release of Midjourney v6, a new arms race for image creation models is about to begin again.