Top 8 neural networks that draw, edit, and enhance images
We have gathered the main competitors of Midjourney and are trying to use their help to have a drawing of an elephant made.
Images generated by neural networks have taken over the internet and become so popular that even the developers of the most well-known one, Midjourney, had to limit free access.
However, Midjourney wasn’t the only one in the game. The Product Star team, along with Pavel Isayenko, the design director at Yandex, compiled a selection of neural networks that can draw based on text prompts and photos just as well, if not better. To ensure a fair competition, all participants were given the same task — to draw a cute elephant.
Here are the winners:
- Kandinsky 2.1 — a free neural network from “Sber.”
- Stable Diffusion — an open-source neural network.
- Scribble Diffusion — generates images from sketches.
- Shedevrum — creates images based on text prompts.
- Craiyon — allows image generation on both PCs and smartphones.
- Dream — edits photographs.
- Image Creator — a neural network integrated into Bing.
- Starryai — draws images based on references.
Now, let’s delve into each one in detail:
Kandinsky 2.1—a free neural network developed by “Sber.”
Kandinsky, developed by “Sberbank,” is a versatile neural network capable of generating images based on text descriptions, drawing images from photos, enhancing existing images, and even blending them together. It is integrated into Discord and operates through the platform fusionbrain.ai. Additionally, there is a Telegram bot, but access to it is limited and may not be available to everyone. If you are not among the chosen few, the bot will suggest switching to Discord or the mobile app called “Salut,” warning that the queue on Discord is significantly longer.
The easiest way to start using Kandinsky is directly through fusionbrain.ai. You will be asked to enter your email and confirm your registration. After that, on the main screen, you will see a prompt: “Enter a description of the image and press the ‘Create’ button.”
You can write the description in Russian. The more detailed your task description is, the better the result will be. Don’t forget to specify the style in which you want the neural network to draw the image. You can describe the style in text or choose from a list by clicking a special button in the bottom left corner. However, the number of predefined styles is limited, while the neural network understands a much broader range of textual descriptions.
Stable Diffusion—an open-source neural network.
Stable Diffusion is an open-source neural network that, when trained, provides users with more tools for self-expression. To fully utilize all the capabilities of Stable Diffusion, it needs to be installed on a computer. The minimum requirements for installation are:
- 25 GB of free hard disk space
- NVIDIA graphics card with at least 10 GB of memory
- Windows operating system
Additionally, there is an online version available that operates through the DreamStudio platform. To access it, users need to have a Google or Discord account for authentication.
This neural network allows users to create images either from photos or based on text prompts in English. While Stable Diffusion also understands the Russian language, the results may be slightly inferior. Users can also choose different styles for their images, such as drawing a self-portrait in an anime style or generating a 3D model of a pet.
By offering these diverse creative options, Stable Diffusion empowers users to explore their artistic visions and bring them to life through AI-generated artwork.
Scribble Diffusion generates images from sketches
The neural network generates images based on your drawings.
To turn your sketches into masterpieces, no authentication is required. Once you click on the link, a page with random scribbles will open. Take them as a basis to explore the abilities of the neural network. Alternatively, you can erase the scribbles and draw your own sketch.
For significantly better results, add a detailed description alongside your drawing. By using specific prompts, Scribble Diffusion can better understand what you have drawn and complement the sketch with necessary elements. Include information about the image style, what is in the foreground and background, what action the object is performing, and so on in your request.
Shedevrum — a model by “Yandex” that draws pictures based on text prompts
The neural network from “Yandex” features a user-friendly interface. To create a drawing, you need to:
- Download the application on your phone.
- Authenticate using your “Yandex ID.”
- Write a text request.
- Wait for your turn in the queue.
On the main screen, the estimated waiting time will be displayed. Until the specified time, you can close the application without losing your place in the queue. When the artwork is ready, you will receive a push notification on your phone.
The neural network only recognizes text-based requests, so drawing an image from a photo is not possible. If you encounter difficulties with your request, a bot will assist you. Simply press the plus sign button, and you will be offered several modifiers to choose from. Another helpful feature is the feed, where all user images and their descriptions are published.
Using this user-friendly interface, the neural network allows users to create artwork based on their text prompts and offers various options to enhance and customize the generated images.
Craiyon — allows generating images both on a PC and on a smartphone
Just like other neural networks, Craiyon draws pictures based on text and understands only English. Generating images from photos is not possible. Before starting your work, you will be offered three styles: art, drawing, and photo. The “None” button allows drawing without selecting a specific style. After that, it’s straightforward: you enter your request and receive several image options to choose from.
You can draw using either a computer or the smartphone app as well.
Dream is a neural network specialized in photo editing
Dream is most commonly used for creating phone wallpapers. It generates images in a vertical format. To create an image, you need to enter a description of up to 200 characters. The neural network offers several dozen styles to choose from, ranging from ink drawings to street art. Some options are only available through a subscription. You can get one free image within the framework of your request.
Recently, Dream has started working with existing photographs. To edit them, describe in your request what specific changes you want. Disclaimer: this feature is in beta testing, so the results may not meet your expectations.
Image Creator is a neural network integrated into Bing, Microsoft’s search engine
Image Creator is a neural network developed by Microsoft. At the end of March, it was integrated into the Bing chatbot. However, you can generate images outside of the chatbot by simply following a link and logging in with your Microsoft account.
The neural network excels at drawing images based on text descriptions, but developers caution that the text on the image may appear unrealistic, and some body parts may be distorted. This limitation is common among many neural networks as new technologies require time to mature and improve.
Image Creator does not offer a selection of image styles to choose from, but it prompts users to specify them in their requests. Users can also indicate the mood and other details of the desired drawing in the request.
Each day, users are given 25 enhancements, which allow them to create images without waiting in the queue. Once the enhancements are used up, the waiting time for a drawing will increase.
Starryai is a neural network that specializes in drawing images based on references
Starryai can draw images based on text requests and existing photographs. After logging in with your Google account, you will have access to five generations per day. If you want more, you can watch advertisements.
To get started, choose the type of AI — Art or Photo. Then, generate your request. If you have a reference image, click the “Upload Image” button, and the neural network will use it as a basis for the new drawing. The “Promt builder” button allows you to select different artistic styles, including textures, painting directions, and names of famous artists.
The free version will provide images in a 4:5 aspect ratio. To change it, you can subscribe. Additionally, the neural network can enhance the image quality for an extra fee.
Once the image is ready, you can continue working on it. Click on the image and press the “Evolve” button. A window will open with the uploaded reference and your original request. This allows you to refine and evolve the generated image further.
Who owns the rights to images generated by the neural network?
At the moment, the answer is: to no one.
The neural network is not a human, so it cannot own any rights. The creators of neural networks do not participate in the generation of images and, therefore, cannot claim copyrights. As for you, the rights do not belong to you either, as you did not play a significant role in creating the image (currently, the idea and text request are not considered active participation).
As there is no regulation in place yet, you can generate as many images as you want and use them as you wish.
Drawing neural networks work well with abstract requests, but they may encounter difficulties when drawing specific details. For example, when illustrating this article, a request was made: “A baby elephant eating bananas, with a city in the background, a sunset, and in the style of Pixar.” Only three out of eight neural networks depicted the bananas, and only two managed to replicate the Pixar style.
Although the logic of creating requests is roughly the same everywhere, each neural network has its own nuances. Therefore, it is recommended to choose one that is most convenient for you and practice writing requests in it. This way, you can achieve more predictable results.