Midjourney V7 internal test chart first exposure, the picture quality fineness pulls full, the effect is stunning and shocking

This week.MidjourneyComing back strong with the new V7. The internal model has been trained, and the scoring system is now turned on, entering the final fine-tuning stage. Netizens have released the birth chart, the effect is stunning, the painting quality fineness pulls full.

GPT-4o is still hot, and Midjourney is probably going to come back with a V7 kill this week.

The latest official blog post sends out a formal invitation to the final V7 fine-tuning scoring party. Immediately following that, a second round of scoring was opened.

Paid subscribers can participate and rate two images at a time for comparison, choosing the best image based on personal preference/taste.

The first round of grading teaches V7 to learn what is 'bad', while the second round of grading focuses primarily on learning what is 'good'.

Some users have already released many images from the ratings, and the results are stunning and shocking.

V7 has a greatly improved sense of detail and resolution, artifacts have been drastically reduced, and the ultimate evolution in character-generating skin detail has been achieved.

Even the cover of the fashion magazine VOGUE can be generated with a single click, with great aura.

The bottom line is that the V7 may be Midjourney's first Omini model.

After V6 came out at the end of 23, Midjourney almost fell silent internally, and was then replaced by FLUX, Grok, Gemini 2.5, and other models that killed each other in succession, replacing the position of the AI Raw Map Throne.

And then recently, GPT-4o "Ghibli" all over the Internet crazy play, people may have forgotten Midjourney, once in the field of AI raw map god-like existence.

Altman warning: want AGI to come quickly, stop using GPU to generate graphs

After more than a year of settling in, can Midjourney get back in the game with the V7?

After a year and a half, 4o still can't replace Midjourney

In fact, the very next day after OpenAI launched GPT-4o, there were rumors online that V7 was about to be unblocked in early April.

In a long post, bigwig Alain Astruc said that the evolution from V1 in February 2022, to V6 in December 2023, took 1 year and 10 months.

And from V6 to today's V7, it's been 1 year and 4 months, taking almost the total time of the first 5 iterations.

The same tip, version V1-V6 generation effect

The pace has slowed down a bit compared to the previous pace, but that doesn't mean it's stagnant.

Alain further stated that he was really blown away by the accurate responsiveness of the new 4o image model, but even so, it still can't replace Midjourney.

He compares Midjourney to a cat with a unique charisma, with a touch of capriciousness, elegance, and even unpredictable spirituality.

It doesn't mechanically execute instructions, but rather opportunities change and surprise, a trait that is difficult to replicate in other models.

It's not like the whole internet is already going crazy with the V7 scoring charts.

It should be noted that Midjourney's official blog notes that these images are not generated by V7 and do not represent the final quality of V7. Netizens interpreted: V7 then released, the effect must be higher than this.

How about, first, a visual feast of demos.

The V7 Raw Graphs are all over the net

Netizen @GenIArt_Fr posted some of his own participation in the scoring system, picture two-by-two comparison of the scoring interface.

His feedback on his initial experience with the V7 was:

Dramatically improved definition of facial details in the background

Cluttering artifacts and inconsistencies are significantly reduced and very few

Better skin texture performance

The realism is impressive

Creator @Rob101Ai has shared a compilation of a video based on Midjourney V7 that brings together all the amazing photo work.

Also, he produced an original song "Don't Stop" with Ukiyo.

This is not only a visual and auditory collision, but also allows the boundaries of human creativity to be broadened again.

@Rob101Ai says it may not matter that people are arguing about which model is better. Instead of obsessing over 'who's first', it's better to accept the fact that AI tools should be allowed to develop their own unique value.

Some people ironically say that if that's what makes the V7 amazing, then good night.

As mentioned earlier, this scoring intentionally showed bad pictures in order for the model to understand what is bad content.

Ultra-realistic character generation, wax-like sensation reduced drastically

This time around, the V7 may have achieved a hyper-evolution in terms of realistic character generation that is simply unrecognizable with the naked eye alone.

Crucially, the new version is a big improvement in the naturalness of the skin, which is much less waxy than it used to be.

See larger image:

In the generation of Asian character pictures, the V7 has also arrived at a realistic rendering.

Other demos for character generation effects:

Creative design, one-click blockbuster

Of course, the real power of Midjourney is the creative design, such as advertising, abstract art, various styles of home design and so on.

Diaper commercials, for one, a Midjourney is enough.

Decorating interiors in a variety of styles.

There are also anime 3D cartoon characters designed to be dorky and cute for less.

Realism, one step closer

As summarized by @GenIArt_Fr above, the V7 is impressive when it comes to realism.

Can you tell if the following live image of a worker fixing a water pipe is not AI-generated?

In addition to characters, the V7 really pulls out all the details in the picture generation of animals and insects.

Generated sandwiches, Michelin delicacies, look very appetizing.

Another person put up some weird picture demos.

For example, six toes, hands full of teeth and various other deformities, and a bloodthirsty horror style ......

Here is a collection of more demos for V7:

It's not the final V7 raw map demo, but these cases are shocking enough.

The throne of the AI Raw Chart may be changing hands again.

ChatGPT surpasses Google, 4o really kills the madness

With 4o's powerful map-generating ability, just last night, ChatGPT search terms, officially surpassed Google. And, "Ghibli" has become the focus of the whole network.

Until now, there are still a lot of people all over the internet who can't stop playing.

In the latest real-world tests, 4o has been able to generate seamless textures/materials and even full PBR maps.

The rise of the AI Ghibli wave, on the one hand, the most debated issue is copyright, on the other hand, is the artist can really be replaced by AI?

AI Ghibli is an insult to art and artists, according to IMAGES' latest editorial piece.

Artists have spent many years to polish their artistic expression and skills, but these efforts are easily plagiarized by a single command of AI, and this kind of "industrialized assembly line generation" is dissolving the value of art.

They criticized that the GPT-4o-generated Ghibli images were superficial, and that the essence of the art was indeed a downgrading of the art, stripping away the true soul of Ghibli's work.

True Ghibli art is never just about dopey characters: from the siblings surviving the war in Grave of the Fireflies to the portrayal of greedy human nature in A Thousand and One, craftsmanship coalesces in every work.

'AI is a tool, not an artist', says hand-drawn animator Usman Riaz bluntly.

However, the real value of GPT-4o at the moment has to be not to replace anyone, but to be a source and script for people's creativity.

Wharton professor Ethan Mollick has given it, great credit - theAI image generationThe 'No Elephants' revolution has been ushered in.

He used AI to generate a room without elephants and labeled it to show why elephants could not exist.

The traditional Microsoft Copilot tool generates the results below left, while the GPT-4o generation is not only accurate, but also follows the instructions precisely.

Prompt: show me a room with no elephants in it, make sure to annotate the image to show me why there are no possible elephants

Or even, a hand-drawn picture of yourself, GPT-4o can straighten out a creative poster.

Home design, refining the pictures one by one according to the instructions.

There's also drone advertising, which 4o has shown to be unrivaled.

Humanity will reach a new creative milestone with the arrival of Midjourney V7.

References:

https://updates.midjourney.com/v7-rating-party/

https://x.com/alanxtruc/status/1905009099013521554

This article is from WeChat"New Zeal."Author: Peach