Midjourney has become famous for its ability to generate images from scratch, but its capabilities are not limited to this.
In mid-November, the fourth version of the neural network was released. The algorithm has learned to create more realistic images, crossing two photos and more accurately processing requests. What is important, we have improved many tools that allow you to generate photos based on a real reference. Now users can style their own photos with high quality.
Social networks have already learned how to generate avatars in different styles. Based on one of the instructions how to add a bit of “Naruto” or “Harry Potter” to your photos.
A little time on Midjourney
This is a neural network that generates pictures from a text description. They can easily be confused with “people” art. Midjourney is the most affordable of the competitors. For Give it there is no public access, and you need a positive computer to run Stable Diffusion.
Midjourney has a large amount of customization. With their help, you can change the details of the picture, aspect ratio and other parameters. We talked more about how to use the neural network in the big guide.
The free version of the neural network has 25 attempts. This is quite enough to generate a dozen high-quality photos. Restrictions can be removed by paying for a subscription, but this cannot be done with a Russian bank card. The second way is to re-create an account in Discord, but there is a risk of getting banned.
How to generate a stylized avatar in Midjourney
One of the Twitter users said in the threadhow to create a photo based on your photo. We recount and complete his advice.
Step 2 In the Newbies tab, go to any Newbies channel. In these channels, users create photos. You will see all their attempts to create a picture using a neural network, and they will see yours.
Step 3 Type the command /settings in the chat and set the following settings: MU version 4, Style med, regular upscale, Remix mode, fast mode.
Step 4 Enter the /imagine command, type any text and send it to the chat. This is a test request that will lead to nothing. After that, a message will appear where you need to accept Midjorney’s terms of use. Click Accept ToS.
Step 5 Choose your photo where the face is clearly visible. Send the picture to any hosting where you can copy the link to the image. Simply uploading a picture to Discord will not work. Postimages hosting is suitable – it does not degrade the quality. Copy the link from the direct link field. The photo can also be copied to Telegram or Vkontakte.
Step 6 Return to Discord, type / imagine again in the chat, and in the prompt line that appears, add a link to your photo.
Step 7 After the space, fill in the picture with a text description of what should be changed. The request will reflect the style of your photo. For example, if you want to make yourself into a Harry Potter character, enter the Harry Potter character. If you like some visual style, add art style, such as Van Gogh art style. It is not necessary to follow these rules, write whatever comes to mind. You can experiment, but remember about the limit of 25 generations.
Step 8 The neural network will generate live image maps. This is not the end result yet. Buttons will appear under the collage – U1, U2, U3, U4 and V1, V2, V3, V4. The first parameter is upscaling. Choose the most successful photos from the cards, and the neural network will improve its resolution and add details at the same time. The second parameter allows you to create new variations of a particular photo in a collage – in case you like the result, but want to see what else the neural network will do. same spirit Free generation is also spent on upscaling.
Step 9 Restart generation by for the same reason request, you can use the arrow button. To save the result, open the image and click “Open Original”. The image will open in a browser tab.
How to achieve a positive result
It is easier to generate from a picture than from scratch: you immediately show the neural network what to work with. But difficulties can arise when compiling a request or when coming up with ideas. Here are some of our tips in case you don’t like the results produced.
Choose selfie or portrait with positive lighting. The figure must be closed. If you use a waist length or full length photo, there is a high chance that Midjourney will ignore your facial features. But in this case, the neural network has room to draw a beautiful and detailed background. So you will probably have to choose.
Specify gender in the request. Midjourney can sometimes generate a male instead of a female and vice versa. Moreover, situations are not uncommon when, for a request, the neural network makes, for example, three girls and a grandfather. If you don’t want to be a bearded uncle, then add girl, girl or woman to the request.
Don’t make a request – take a picture. The neural network can blend two images. Instead of writing a request, you can simply copy the link to any image with a character or style you like. This is how the neural network copes even better: several times I failed to generate myself in the image of a superheroine through a request, but it worked perfectly with the picture.
Change the pictures. Some photos come out better with an animated style, while others come out with a more realistic style. If a photo doesn’t work in a certain style, try a different photo.
Try to convert photos with animals. We especially recommend cartoons like “My Neighbor Totoro”. The neural network very much likes cat and dog style.
Use a special service for selecting a request. If you can’t figure out what style to get yourself into, use prompters – small services that help formulate a request. There is a list of popular styles that the neural network is positive at dealing with. We recommend Prompter and Phraser.
What other neural networks can be used to process pictures
Dream Studio. The light version of Stable Diffusion Neural Network has the ability to upload your photo and add a investigation on it. It seems not as high quality as in Midjourney, but the neural network adds more details to the request. Up to 200 images can be generated for free.
Draw Anyone. The service works on the basis of stable streaming and generates avatars in the styles set by the developers. Need to download 5-10 clean photos and wait while they are processed. The mat is long – the process can take up to four days. Only some images can be saved for free.
FaceTune. Added to the face retouching application AI generator stylized image. You need to upload five of your photos and wait until they are processed. After that, you can either apply the filters prepared by the developers or write your own query. You can generate images for free and without limits. The app is available for iOS and Android.
We post circles, beautiful cards and news about technology and pop culture in our telegram channel. Subscribe, it’s positive here: @t_technocult.