In 2021, OpenAI introduced the worldneural network DALL-E, which can create images from a text description. From that moment on, services that can perform the same function began to appear like mushrooms after the rain - earlier we even made a selection of the best ones, here is the link. It’s the end of 2022, and today the number of neural networks that draw, according to the words, has increased significantly, it’s impossible to keep track of their appearance. And it is already difficult to guess which of the images published on the Internet is drawn by the artist, and which is the result of the work of the neural network. In the framework of this article, we propose to learn about neural networks-artists, the existence of which many do not even know. No ruDALL-E and the almighty Midjourney - only rare examples.
Content
- 1 Neural network for creating anime - QQ Small World
- 2 Starryai neural network for creating paintings
- 3 Pixray service - a constructor of several neural networks
- 4 Hotpot - a neural network with different drawing styles
- 5 This Does Not Exist - Generation Of Humans And Animals
Neural network for creating anime - QQ Small World
For a while, this Chinese neural network waslittle known, but at the end of November 2022, her images flooded the Internet. It allows you to turn anyone into an anime character in a couple of clicks - we tried it, and it even processes photos with animals, look at the image below.
Using the QQ Small World neural network is simple, even despite the design shining with all the colors of the rainbow with a bunch of animations and Chinese characters.
First of all, you need to open the site - it works both on a computer and on smartphones.
Next, you need to take three simple steps:
- click on the start button of the neural network;
- choose a photo (best if a person can be seen from the waist up);
- after the video showing the changes, press the left button (blue) to get the before/after image.
Here are some examples of images taken with QQ Small World.
Starryai neural network for creating paintings
The previous neural network, although little known, is already at the peak of its popularity today. But little is said about the Starryai neural network - after trying it out, I understood the reason.
To start using Starryai you needSign in to the site using a Google account. After that, there are three models to choose from: Altair, Orion and Argo. For example, I chose the Orion neural network, chose the Lovecraft style, and gave the command to create a “monster dog with wings on Mars.” The process took about 5 minutes, but I didn’t like the result - the first pancake was lumpy.
After that, I tried my luck with a neural networkAltair, having tried to create "mushroom trees" in the style of Unreal Engine. In advanced settings, you can choose the image orientation and several other options. Unfortunately, as much as I would like to see the result no worse than Midjourney, what you see below turned out.
Perhaps the Starryai neural network also does not usepopular because each launch costs 2 "credits". They can be bought for money or earned by watching ads or sharing the results of your work on social networks.
Of the advantages of Starryai, one can only single outthe ability to look at the progress when creating an image and save the result as an animation. As for me, this neural network can be launched only for the sake of interest - maybe you can do something better? Send the results to our Telegram chat, maybe among our readers there is a master for compiling queries for neural networks.
See also: The most impressive paintings, music and other projects created by the neural network
Pixray service - a constructor from several neural networks
If you want something free and withoutrestrictions - the Pixray site may be suitable for you. This is not a single neural network, but a small service with several popular open source models for creating pictures from a text description. I tried to give the VQGAN neural network the task of drawing "a monkey in a Santa suit riding a lion". See the result below - the quality is so-so, but you can distinguish the common features of a monkey, Santa and a lion.
Unlike other similar services,Pixray allows you to enter settings - there is a small instruction on their use and examples in the "Examples" section. Since there are no restrictions and no payment, you can play enough. But you shouldn't expect masterpieces like Midjourney's.
Hotpot - a neural network with different drawing styles
Unlike the above options,The Hotpot neural network produces very good results. To create small pictures, it is free - you need to pay only if you want to receive images in large resolution. To show examples, I tried to make Johnny Depp in a space suit and Elon Musk as a noir detective - in my opinion, the result is good. The only downside is the low resolution of the free images. The Hotpot neural network fits perfectly as a source of inspiration.
This Does Not Exist - generation of people and animals
In 2019, we already talked about the site ThisPerson Does Not Exist, which on each update generates the face of a non-existent person. Recently, I was surprised to learn that there are sites that generate cats, sneakers, and even resumes. For many, this can also be a big discovery - for a few minutes, these sites will definitely captivate.
Here are some more articles about neural networks that can create images:
Also, don't forget to subscribe to our Zen channel, where you will find many interesting articles on science and technology!
Perhaps you know about other interesting neural networks that work with images. If yes, write in the comments.