Eight months after its last model update, Midjourney has unveiled the seventh version of its AI image generator. But what are its strengths and weaknesses?
More than a year after the release of v6, Midjourney has unveiled the seventh version of its AI-powered image generation model. A heavyweight in the field, the publisher offers a new version that shows impressive potential, emphasizing its arty side but struggles to mask its experimental nature, with many flaws remaining.
How to access and test Midjourney v7?
Midjourney is an AI that no longer offers free credits to experiment with, or on rare occasions. To use it, you will need a paid account, then access the web interface, the tool does not yet have an official mobile application. Once this is done, you will discover, or rediscover, the classic interface of the platform and this type of generator. To understand it, you can refer to our practical guide. Under the dialog box for writing prompts, after clicking on the Settings button, you will notice that version 7 is now the model selected by default.
However, there is currently a small subtlety: to be able to generate images with version 7, you will first need to create a personalized template. To do this, Midjourney invites you to activate template customization and then, quite simply, like an image from among two proposed. This will allow the AI to create a profile for you with your preferences. 200 likes are required to create your profile, but the more you refine this customization, the more it will be adapted to your artistic preferences. Once this is done, your profile is created, with a unique code: these codes allow you to use other styles created by other users, an option previously available but not mandatory, and which remains so for older models.
What are the advantages of Midjourney v7?
As the “smartest, most beautiful, and most coherent model to date,” according to Midjourney, this v7 has “unique strengths, and probably some weaknesses .” After our first tests, here are the observations we can make on the qualities and defects of this latest version.
Often stunning photorealism
Midjourney was one of the first AIs to impress with the realism of its generated images. This seventh version is no exception. With the right prompts— you can quickly achieve convincing photorealistic visuals.
- A domestic cat curled up asleep on a white windowsill, sunlight streaming through sheer curtains, soft shadows on the wall, warm indoor tones, fine fur detail, peaceful and realistic vibe, captured with a 50mm lens.
- Wide-angle shot of a dramatic mountain landscape bathed in golden hour light, soft haze in the distance, long shadows across the valley, rich warm tones, captured with a 24mm lens, f/8 for deep depth of field, high dynamic range, ultra-realistic detail.
- A street vendor in a bustling Asian market, skillfully preparing traditional cuisine at a food stall. Photographed with a Canon EOS R6 using a 35mm f/1.8 macro lens at f/2.2 aperture, 1/200 second shutter speed, ISO 500. The image captures the steam rising from the cooking pot, the vendor’s focused expression, and the colorful array of ingredients.
A perfect imitator of graphic styles
In recent weeks, controversy has swelled around OpenAI and ChatGPT’s astonishing ability to almost perfectly imitate the style of Japanese animation studio Ghibli. Less generally accessible, Midjourney has rather slipped through the cracks, even though its models have long been adept at copying artistic styles. This version 7 is no exception, as the examples below demonstrate:
- A lone traveler in a pastel desert with surreal rock formations and giant structures on the horizon. Imitating Moebius’ comic book art style, clean line art, wide compositions, vivid yet dusty colors, mystical futuristic ambiance.
- A woman sitting alone in a small diner at night, warm interior light contrasting the dark empty street outside. In the style of Edward Hopper, minimalist composition, muted color palette, cinematic realism, subtle tension.
- A bustling open-air market in a mountaintop town, filled with eccentric merchants selling magical items, talking vegetables, and flying books. Children run between stalls. Clouds drift below the cliff’s edge. Ghibli-inspired, bright colors, rich environmental storytelling, warm and lively atmosphere, anime-style linework.
- A contemplative woman in profile, surrounded by swirling patterns and ornate gold leaf textures. Styled like Gustav Klimt, detailed ornamentation, mosaic backgrounds, elegant color scheme, sensual and symbolic.
A rather inspired and dreamlike AI
Midjourney is, if you will, a rather inspired and creative AI. It often offers original results, with a rather characteristic arty touch. This sometimes offers beautiful compositions, imbued with a certain dreaminess, when we go further than the search for cold realism (which Midjourney is also capable of producing).
- A stylized vector scene of a rocky coastline at sunset, flat shapes and bold colors forming cliffs, sea, and sky in pleasing contrasts.
- A bustling city street illuminated by neon signs reflecting off the wet pavement after a heavy rain. Photographed with a Sony a7 III equipped with a 35mm f/1.4 lens at f/2.0 aperture, 1/60 second shutter speed, ISO 800. The scene captures pedestrians with umbrellas, the motion blur of passing cars, and the vibrant colors of the urban nightlife.
- A girl looking out of a quiet train window, alone in the carriage, farmland passing by outside. Inspired by Hayao Miyazaki’s hand-drawn anime style and Edward Hopper’s composition, soft light, emotional stillness, painterly backgrounds.
What are the flaws of Midjourney v7 when it was released?
This new version of Midjourney has been highly anticipated. However, it may disappoint some users, initially, as it is not without its flaws.
The return of imperfections
Poorly maintained objects, inconsistencies, imperfections… This first draft of Midjourney v7 has a few glitches. Here, a pedestrian walking in the middle of the street, there, a traffic light that doesn’t belong there, or even here, a slightly creepy face… These small errors are less visible in version 6.2 of the tool and recall some of the flaws of version 5.
- A bustling open-air market in a mountaintop town, filled with eccentric merchants selling magical items, talking vegetables, and flying books. Children run between stalls. Clouds drift below the cliff’s edge. Ghibli-inspired, bright colors, rich environmental storytelling, warm and lively atmosphere, anime-style linework.
- A bustling city street illuminated by neon signs reflecting off the wet pavement after a heavy rain. Photographed with a Sony α7 III equipped with a 35mm f/1.4 lens at f/2.0 aperture, 1/60 second shutter speed, ISO 800. The scene captures pedestrians with umbrellas, the motion blur of passing cars, and the vibrant colors of the urban nightlife.
When the detail is too much
A prompt that’s too complex and teeming with details will cause Midjourney some difficulties. Indeed, the longer and more specific your query, the more difficult it will be for the tool to logically integrate certain elements. Therefore, it’s best to initially limit yourself to prompts that include relatively few different objects, characters, and actions.
- A forgotten kingdom thriving after a silent apocalypse, baroque palaces covered in overgrown bioluminescent plants, mutated deer with crystal antlers grazing near golden fountains filled with stardust, the sky cracked open with aurora leaking through, dreamlike atmosphere, mix of fantasy and decay, highly detailed environment, dramatic lighting.
- A cybernetic oracle floating in a data ocean, her robes made of shifting code strings and translucent quantum threads, surrounded by holographic whales made of forgotten internet data, ancient circuitry engraved on obsidian monoliths behind her, glitchy yet sacred atmosphere, spiritual cyberpunk aesthetic, vibrant digital colors.
Text generation at bay
GPT-4o surprised everyone with its ability to reproduce text identically. Reve and Ideogram also demonstrate excellent skills in this area. As for Midjourney, we’ll have to forget about it. At least for now. Integrating and reproducing coherent text are clearly not among the model’s strengths. Errors are common, even for short texts…
- A stylized 1980s-inspired album cover, neon shapes, a figure dancing under a disco ball, deep blues and magentas. Bold, large title text: “ELECTRIC LOVE” in retro chrome typography, artist name: “Miki Voltage” in cursive below. Inspired by synthwave aesthetics and Japanese LP designs.
- A dramatic movie poster of a young man standing under neon lights in a rainy city street, cinematic angle, bold lighting contrast, moody colors. Title at the bottom in bold sans-serif font: “LONELY NIGHTS”. Subheading: “A Film by J. Nakamura”. Inspired by modern Japanese indie cinema, realistic typography placement, minimal poster design, high-resolution.
A promising v7 but largely improvable
This v7 of Midjourney will undoubtedly disappoint those who were expecting a revolution in AI image generation, after the sensational arrival on the market of the new OpenAI model integrated into ChatGPT. It must be admitted that the “wow” effect is not really there, even if Midjourney has not lost its qualities developed over the years. This seventh version seems very experimental and, like the previous ones, we will probably have to wait for version 7.1 or 7.2 for the platform to acquire a more complete tool. Similarly, the editing options are still limited, proof that this model is still in development.
However, it’s also important to note the importance of personalization. The mandatory ranking phase before using the template demonstrates that this aspect is particularly important. The more you refine your profile by liking images, the more Midjourney will generate visuals that reflect your tastes, preferences, and desires/needs. A word of advice: spend some time ranking images—potentially up to a thousand—to get convincing results more quickly. In any case, Midjourney offers great promise with this version 7, which will need to be confirmed in the future.
