Best neural networks for making creative pictures and videos
08 Jun
2023

How to create pictures for posts and landing pages without designers, and videos without actors and cameramen? Of course, using the neural networks, which produce the right visuals at the snap of a finger. In our previous articles we've already looked at ChatGPT and how to use it in an affiliate program, and made a compilation of generators of unique texts to attract traffic. Now let's take a look at the best AI-based services that will draw cool pictures and edit videos for your content marketing.

Midjourney – No. 1 in creative imaging

The Midjourney neural network is the most popular image creation service based on a text request. The neural network has already taken part in developing ad campaigns for Heinz, Coca Cola and other brands. Some of CleverAff's partners also use Midjourney to fill social networks, make creatives for ads, images for landing pages, etc. So we decided to disassemble the whole process of working with the service and share some tips to make it easier for you to implement this tool.

Midjourney offers to generate 4 image options for free. For money, three subscription packages are available, in which the user is given time to generate images – from 3.3 to 30 hours per month.

Step 1. Registration. To get to work with the neural network, sign up or sign in to your Discord account (free messenger) and accept the invitation to join the Midjourney server. In the left panel of your messenger, the neural network logo will appear – this is the server. 

Pic.1. Where to find the Midjourney server

Step 2. Choose a chat room. On the server, open any of the chats called "Newbies". Interesting fact that in the chats you can see the creatives made by the neural network based on requests from other users.

Pic. 2. Selecting a chat room where you can use Midjourney's neural network

Step 3. Setting up Midjourney. To configure the basic parameters of the generating, type in the text box with the command "/settings". The bot will send you the settings options in response.

"MJ versions" are versions of the engine, which the developers are constantly updating. By default, it is the latest version. If you want a picture in anime style, then select the engine "Niji version". The parameter "Style" is the level of stylization. The higher it is, the more creative the neural network will produce, but the result may be far from the request. The next parameter is the access mode. In the trial version you can use only "Public mode", which shows the result of your request to all users of the server, and on subscription – "Stealth mode". Next you have to configure the generation mode. The default is "Fast mode", which generates the image in 30-60 seconds. "Remix mode" is a mode of combination. To use this function, you need to upload links to 2 images from public sources.

Pic. 3. Image generation settings

Step 4. Getting started. To create your own design, send the chat command "/imagine", which opens a field for entering a request.

Pic. 4. Command "/imagine" and the input window of the request 

Step 5. Request. Making requests for Midjourney is a special kind of art. A basic request is a word, a short phrase, or even an emoji. Every word and symbol affects the final result. Use these tips to make a quality request:

 

  • write in English so that the neural network understands the task correctly;
  • describe the basic idea first, then add the details;
  • use adjectives to describe the shape, color, size, and object;
  • give clear instructions: instead of "lively cityscape" it is better to write the name of a particular place, for example "Times Square";
  • avoid ambiguous phrases, abstract descriptions, and the "not" particle;
  • detail the composition: mention the position of the subject, style, lighting, atmosphere, and camera type, for example: Kodak Gold, Agfa Vista, Canon EOS 5D, Kodachrome, etc.

 

Midjourney does a great job with short requests, but you can improve the result if you add special parameter setting commands to it. They are specified in the request with a "--" (two hyphens) in the format "--command X", where "X" is a numeric value or word. Let's look at the most popular ones: 

 

  • --v – the version of the engine (1, 2, 3, 4, 5, or 5.1);
  • --s – stylization of the image, the higher the value, the stronger the abstractness is expressed (range from 625 to 60000);
  • --ar – specifies the aspect ratio (the default is 1:1, you can choose another, for example, 16:9);
  • --q – the quality parameter which adjusts the clarity of the drawings (range from minimum quality to high 0.25-2);
  • --wallpaper – to create a detailed picture in desktop wallpaper format.
  • --chaos – adjusts how different the resulting versions of the image are, the higher the number, the more different the pictures will be (the default is 0, but you can change to any value up to 100);
  • --no – removes some element of the already generated image (for example, " --no trees" or "--no people");
  • --video – saves the process of creating an image in clip format.

 

One last tip for making a request in Midjourney: separate the parts of a long request with "::" (two colons). The formula looks something like this: "Object description :: composition details (stylistics, lighting, atmosphere, camera, etc) :: short commands (for example, --q 2, --ar 16:9)".

Lifehack! To avoid the hassle of writing a request by hand, you can use services that will do it for you: Promptomania, Prompt helper, etc.

Step 6. Creating an image. After all the manipulations with settings and description you send a request to the bot. As a result, it will give out 4 previews, of course, if no other number is specified in the settings. For example, we asked Midjourney to generate a picture of a trader at work.

Pic. 5. Example of generated images with a trader

Step 7. Refining and saving. After the neural network has given you the result you should save the program or continue working with it. To save the pictures you like, press the button with the "U" marker and the preview number. If you press the "V" button, the bot will create another 4 versions of the preview.

For example, we like the picture number 2. But the result is not perfect yet, so we press "V2" and get 4 more options. 

Pic. 6. Generating new images from preview no.2

Additional features. Besides generating new images, Midjourney can also work with references. For example, if you send us a link to your own image or any image from the open source and specify the necessary parameters, then the neural network will generate a modified image. For example, we decided to generate a screensaver for YouTube based on our reference.

Pic. 7. Image of the price chart as reference

In the request input box, we left the link and added the "--qr 16:9" parameter. Since Midjourney doesn't understand what a screensaver on YouTube or a picture for a Facebook post is, it's important to specify the right size of the picture right away.

Pic. 8. Generated image by reference with "--qr 16:9" parameter

In Midjourney, you can adjust the level of influence of the original image on the result. With a short command " --iw" we set how close or how far you want the result to be from the reference (the default is 0.25 and the maximum is 5). For the YouTube screensaver we had earlier, we specified the parameter "--iw 0,1", and this is what we got:

Pic. 9. The resulting images with the "--iw 0,1" parameter

A few more examples of Midjourney's short request images for social media posts.

Pic. 10. Picture for the post about trading strategies in the comic book style

Pic. 11. Picture for the post with a selection of movies about trading in the style of the movie "The Wolf of Wall Street"

Pic. 12. A detailed picture of the trader

Pic. 13: The picture of a bear and a bull in the cartoon style

You can see from the last result that the neural network can produce errors. Only one of the four previews turned out to be a normal bear and bull, while the other versions showed strange animal hybrids. Yes, Midjourney is not a perfect neural network, but it is good enough to handle most simple design tasks.

Other image generators

In addition to Midjourney, there are other neural networks that can help you to create quality images to attract traffic. Many of them work for free.

  • Lexica Aperture is a free service that creates pictures in a realistic style for short requests. The neural network does not require serious settings, just describe the idea, choose the size, the version of the engine and the "Guidance skale" level (how accurately the neural network should visualize the request). In settings there is also a field for entering a "negative request", that is, something that shouldn't be in the picture. The service can also work with references.
  • Dream by WOMBO is a simple generator that creates images based on the description and choice of style. You can upload your own pictures, which the neural network will transform into something new. In a free version, the service gives you only one picture at 960 x 1568 pixels, and with a subscription you get 4 variants. Interesting fact: Dream by WOMBO allows you to edit the result with text.
  • Another free service is Stable Diffusion. Like the previous one, it has a minimal set of settings: a field for entering a request and a scale to match the text description. Stable Diffusion generates four picture variants of 512x512 pixels.
  • Bing Image Creator is a Microsoft development that is thought to be a worthy alternative to Midjourney. The service is free and available from any browser, but in Microsoft Edge it is already built into the toolbar. Image Creator generates beautiful 1024 × 1024 pictures by description. The user gets 25 quick generations at the start, after using them the creation will be much longer. Fast generations are resumed daily. The neural network can be accessed after authorization with a Microsoft account. Since in some regions the ability to use Microsoft products is limited, a VPN may be needed.

Video and voice generators

Video content is extremely important for attracting and warming up your audience. Commercials, informational and training videos for Youtube, stories – you can create all that using neural networks. 

 

  • Multifunctional Designs.ai service generates and gives voice to video clips based on stock content. The arsenal of neural network includes: 10 million videos, 170 million pictures and graphics, 500 thousand pieces of audio. To generate a video in Designs.ai, you need to make a script of up to 2.5 thousand characters and upload it to the request window, select the industry, language (over 20 languages, including Russian) and voiceover. After learning the script, the artificial intelligence automatically selects appropriate video files, provides voiceover, and adds music and titles. This is the video generated by the neural network according to our script. You can customize the final result: change the video, titles and audio, add transitions, upload your own logo and other files. The service also lets you optimize your video for any social platform. Editing is limited in the free version Designs.ai, but to fully experience the service, sign up for a 30-day trial.
  • InVideo is a simple and straightforward video content creation service that works based on templates. To create a video, you need to choose a format, a suitable template and write a script. By the way, InVideo can also help you with the script. Artificial intelligence will generate a script for each scene based on a short request and keywords. The service can also convert text to voice and animate pictures. You can use the tool for free, but with some reduced capabilities.
  • Zvukogram is a pocket-sized speaker, who will voice anything with a beautiful male or female voice. The library of voices has about 50 variants, among which there is also an imitation of a robot's speech. To get a voiceover, you need to compose and upload a text in the request window, select a language, a speaker, the pitch of the voice, and the speed of narration. Life hack: place stress and punctuation in the text, so the neuronet will read the words with the right intonation and pause in the right places.
  • The program Faceswap is useful if you need to replace the face of a character in the video to another. Faceswap is based on DeepFake, and it produces very good results. To create such a video, you have to install the program on your computer, upload the video and choose the face from a huge library or add your own file. The program also knows how to remove the background from the video, which allows you to get a completely new picture.

 

Neural networks are designed to simplify complex tasks. Previously, it took a lot of time and money to write an advertising text or article, draw a picture or mount a video, but now artificial intelligence does all this in a matter of seconds. Using the services that we have reviewed in this series of articles, you can quickly generate as many texts, pictures and videos as your traffic acquisition strategy requires. And if you need branded promotional materials for work, contact the CleverAff manager.