Midjourney v5 is now available. The new model can generate much more realistic and detailed images. We explain all this to you.
If you follow Frandroid diligently, you’ve probably noticed images designed thanks to Midjourney. We use this tool to generate artwork. A fifth version was recently launched, featuring a significantly reworked neural architecture and new aesthetic approaches. While GPT-4 gets a lot of media attention, Midjourney is another tool that also deserves our attention and should be closely monitored.
What is Midjourney?
Midjourney is an independent research laboratory that has developed an eponymous program based on artificial intelligence. This program is designed to generate images from textual descriptions, such as OpenAI’s DALL-E and Stable Diffusion. While the underlying technology appears to be Stable Diffusion, Midjourney’s creators have developed their own unique style, with an in-house algorithm, that has attracted a lot of attention.
Midjourney’s V5 has arrived and it’s finally doing five-finger hands 🖐️ pic.twitter.com/euP2VkKWfm
— Numerama (@Numerama) March 16, 2023
The tool has been in open beta since July 2022. Since then it has been attracting more and more users: version 2 of their algorithm was launched in April 2022, followed by version 3 in July of the same year. On November 10, 2022, the alpha iteration of version 4 was released to users. Finally, Midjourney in version 5 has been available since March 15, 2022.
Midjourney works with simple words, but also uses “prompts” – advanced commands – to create detailed and accurate images with the Midjourney bot. This fast provide a textual description of the image you want to generate and guide the bot on how to get to it. Whether you want a photo, a painting or a quick drawing, the fast can help the bot create the exact image you want to achieve.
What changes in this fifth mountain?
Midjourney creator David Holz describes V5 as a professional tool because, unlike previous versions, it is not limited to specific art styles and offers a wider variety of visual results. However, this V5 is still in the alpha stage and will undergo “major changes” before the final release.
In short, it gives more creative freedom, but this freedom must be tamed. According to David Holz, this may have to do with usage fast more elaborate details that explicitly describe aspects of the image, such as lighting and atmosphere, to achieve the desired result. Note in passing that it is important to specify precise details so that the bot generates specific elements. Think of subject, medium, environment, lighting, color, atmosphere and composition. Anything not mentioned is randomly generated.
After trying V5, we saw more accurate, more relevant results, and fewer anomalies or glitches. However, image creation is slower. This situation is similar to what can be observed with GPT-4 on ChatGPT. The V5 version offers significant improvements in language comprehension, accuracy and aesthetic flexibility, all of which contribute to the generation of breathtaking images. The depth and richness of the images he produces are truly astonishing, and the range of aesthetics he can conjure up is astonishing.
How do you use Midjourney V5?
Currently, MidJourney is only accessible through Discord (take the opportunity to join our waiter), so to start using it, you need an account on Discord. First, visit the MidJourney website and click on the Beta Participation Program.
After that, you need to accept an invitation to the Discord Midjourney server. Now open your Discord app and click on the MidJourney server boot icon. Use the prompt with the command /invented and start creating images from your imagination. You can do it on the server you want, obviously you will have to configure Midjourney in the latter.
To activate V5 you need a paid account, then activate this version with the command /institutions. Note that V4 already works very well, but a free account will be limited in the number of possible requests.
How to use prompts?
Basic commands are as simple as a single word, phrase or even an emoji. The bot doesn’t understand grammar or sentence structure, so word choice is important. More specific synonyms work better than generic synonyms, while using fewer words gives each word more weight. You also have the option to use fastthe famous advanced commands.
Advanced commands can contain one or more image URLs, several text sentences, and one or more parameters. Image URLs can be added to a prompt to influence the style and content of the final result, and fast can change the way an image is generated, such as proportions, models… Numerama cites several examples to start well. Moreover, this site generates fast on the run.
It’s also essential to focus on what you want in the picture rather than what you don’t want. Finally, speak to Midjourney in English for more accurate results. If this is not your native language, you can use DeepL for translations.
Some examples generated with Midjourney V5
Here are some additional examples of images generated with Midjourney V5.