2022 was a breakthrough year in the field of neural networks. Then the famous ChatGPT from Open AI and, of course, Midjourney, a neural network for generating images, came out. Of course, there were image generators before, but not a single neural network showed such a level.
This event has influenced many industries that require a lot of creative ideas and a constant flow of images. Now the world, without exaggeration, will not be the same, and in vacancies you can increasingly find a requirement for “AI skills” or “willingness to work with neural networks.”
Because to continue to be a highly qualified specialist you need to learn new tools. Today in this article we will look at the ten most popular image generators and talk about their pros and cons.
What is an AI Image Generator
An AI image generator, as clear from the name, generates images. It’s important to emphasize the word “generates” because some AI chats may search the internet for images instead of generating them, while neural networks for generation create new, unique images.
Such images can be guaranteed for commercial use, but it’s important to note that in many countries, legislation is considering regulating issues related to neural networks. One pressing issue is the ownership of the work created by AI, especially when the network learns from existing works of authors and generates images based on them without creating anything of its own.
Many text-based chats for communicating with AI work in conjunction with image generators. For example, DALL-E can be used in communication with ChatGPT. In this case, we will specify in the article where the generator can be used.
To generate images, neural networks need a prompt. A prompt is a textual description of the image you want to obtain.
Since artificial intelligence is still quite primitive, most neural networks understand technical simplicity well – sets of special symbols and words in the form of commands, written through “;” with characteristics of operation. For example, the style name in which the image should be, its proportions, camera settings, preferred colors, etc.
We could talk a lot about prompts, but today we will only evaluate the ease of communication with the neural network.
Midjourney
In 2022, unexpectedly for everyone, Midjourney was released, which is currently the best neural network for image generation. Previously, all neural networks for working with art only performed separate functions: adding an object to a photo or removing it, stylizing the image in different styles, aging people in photos. Midjourney, on the other hand, is a new level; with this network, you can generate an image completely from scratch based on textual description.
The image below was created with the prompt “generate an image in the style of Van Gogh’s paintings with a girl in sunglasses.”
Capabilities and Ease of Use
The neural network is constantly updated, making it difficult to track what the current version of Midjourney can do. An active community of specialists from various fields experiments with the generator’s capabilities, and you can find many ready-made prompts for work or recommendations and even training courses on working with this network online.
This abundance of information makes working with the neural network easier.
It’s worth noting that even without an active community, the neural network is quite easy to work with, and the developers’ website has a lot of educational content with examples of possible prompts, so even a beginner can understand.
Suitable for all professions without exception.
Image Quality
Midjourney is loved for the high quality of generated images and extensive options for adjusting the camera, canvas proportions, and even color and object proportions in the finished art.
The problem inherent in all neural networks today with oddities and repetitive elements is also relevant for Midjourney, but with each version, it becomes less noticeable.
There are, of course, errors such as three legs on one person or ten fingers on one hand.
Below, we generated an image with the prompt “generate a homeless person with a laptop and a ginger cat.” There are some issues with the fingers, and the cat is not ginger for some reason. But overall, the image is decent even with a simple and imprecise prompt (Midjourney prefers technical descriptions-commands and poorly understands natural human language).
Performance
Due to high server load and the complexity of the work performed, the neural network consistently lags. This is something that must be accepted, as there is currently no equivalent to Midjourney in terms of quality.
Pricing
Initially, Midjourney only worked through a Discord server, and each user had twenty-five free prompts. But now there is no option to use the neural network for free at all; you must purchase access immediately.
Pros and cons
| Pros | Cons |
| Active community, ease of learning | Distinctive traits in style that make Midjourney’s authorship easily recognizable |
| High level of generated image quality | No free trial version |
| Extensive camera and proportion settings, ability to control every detail the neural network generates | Operates slowly |
Leonardo AI
Capabilities and Ease of Use
Working with Leonardo AI is easy due to the convenient website interface. It understands human language but behaves strangely during generation. For example, it doesn’t generate anything extra: if in the examples above the neural network, when generating a homeless person with a cat and a laptop, guessed that if the person is homeless, then he probably sits on the street and looks accordingly, Leonardo AI generated a homeless person with a laptop and a ginger cat inside a house in response to the request “generate a homeless person with a laptop and a ginger cat.” Below are some results.
Therefore, to work with this neural network, you will need to carefully study the prompt; otherwise, the result will be unexpected. For example, in the prompt, it was mentioned about a ginger cat, but Leonardo AI persistently generates both a ginger cat and a person.
Image Quality
The quality of the images leaves much to be desired. Of course, with the help of accurate prompts and carefully chosen words, you can achieve a normal result, but so far, each work clearly follows one style, and there are problems with generating faces and features.
Below is an image generated with the prompt “generate for me a homeless person sitting next to a garbage bin on the street, with a laptop in hand. A red-haired cat is sitting next to him, against the backdrop of a metropolis.”
Only by refining the result to the maximum did we get an image comparable to examples from previous neural networks. The main problems of the neural network are well noticeable here: sausage-like hands, crooked faces and features, clumsiness of the image. Interestingly, the person is again red-haired along with the cat.
For comparison, let’s see what will happen with the request “generate an image in the style of Van Gogh’s paintings with a girl in sunglasses.”
Performance
Images are generated on average within ten seconds.
Pricing
Advantages and Disadvantages
| Advantages | Disadvantages |
| Free to use | Low image quality |
| Generates exactly what you ask for, without making up anything “from the head” | Generates exactly what you ask for, without making up anything “from the head” |
| Convenient interface with settings for image size, proportions, and other parameters | The free version has strict limitations on the number of generations per day |
Overall Conclusion
The generator struggles with open-ended tasks, but it has immense potential and learning capabilities. For example, for gaming, you need game props – chests, coins, textures. Show Leonardo AI images in the desired style, use prompts skillfully, and you get a very decent result.
Deep AI
A primitive neural network that generates images in a strange way. The result is something like collages made by children in kindergarten during a creativity lesson. You can laugh at this and forget it like a bad dream.
Capabilities and Ease of Use
Working with the neural network is easy, but achieving any decent result will require titanic efforts.
Image Quality
Let’s just take a look at the request “generate for me a homeless person sitting next to a garbage bin on the street, holding a laptop. A red cat is sitting next to him, against the backdrop of the metropolis.”
Issues are not due to an incorrect prompt. If you look at the gallery (not recommended at night, for pregnant women, and children) of images on the website, you can see that all images are equally nonsensical.
Speed
Instantaneous generation.
Cost
The free version allows access to all basic features of the neural network. Specific features in PRO for $5 per month are unknown.
Pros and Cons
| Pros | Cons |
| Can have a laugh | Very low image quality |
| Partially paid |
Canva AI
Canva, a leading tool for creating creatives in design, branding, and other industries, has released the Canva AI tool.
Features and Ease of Use
Now, in the PRO version, alongside other tools, a field for entering a text prompt has appeared in the left panel of the site when creating a creative. Canva AI understands natural language, so there is no need to learn a prompt for a good result.
The generator is designed for one purpose – to speed up and improve the process of creating design content, making it suitable exclusively for these purposes.
Image Quality
The image quality may not always be as expected, but considering that Canva AI was released recently, this is forgivable. Below are examples of images generated by this AI.
Speed
Generation is almost instantaneous.
Cost
Canva AI is included in the PRO subscription in Canva.
Pros and Cons
| Pros | Cons |
| Good tool for designers | Paid |
| One-month free trial version available |
NightCafe
NightCafe is a tool with very broad capabilities for image generation. It combines many advanced neural networks and allows creating your own styles based on them.
Capabilities and Ease of Use
There is a substantial community around NightCafe, making it easy to ask interesting questions, seek help with prompts, etc. The generator’s system encourages cohesion by providing additional credits when images are published in the community.
NightCafe has a very user-friendly interface and generation system, making it suitable for beginners as well.
Quality of Images
Comparing the quality of images with advanced neural networks is currently challenging. However, it is believed that NightCafe will soon catch up with its competitors and provide strong competition. The generator has everything it needs: a cohesive community, a unique offering of multiple neural networks within one project, a system for independently creating styles, and the ability to publish prompts on the website for others to use.
Even now, it can’t be said that everything is bad. But to achieve good results, it will definitely be necessary to carefully study the generator and find the most effective way of interaction.
The first image was created with the prompt “create an image for me in the style of Van Gogh’s paintings with a girl wearing sunglasses.”
Creating an image with a homeless person and a laptop turned out to be much more challenging. Most likely, AI still struggles to understand natural language, despite claiming otherwise, hence the difficulties. However, if we overlook the cat’s face instead of a person’s face and similar errors due to prompt inaccuracies, the work looks more than satisfactory.
Generation prompt below: “create a homeless person with a laptop and a ginger cat.”
Generation prompt below: “create a homeless person sitting next to a trash bin on the street, with a laptop in hand. A ginger cat is sitting nearby.”
Performance
10-15 seconds.
Pricing
A free version is available with its limitations. The number of uses in the free version is restricted, but additional credits (images are generated per credit) can be obtained by publishing generations in the community and other actions.
Pros and Cons
| Pros | Cons |
| Ability to create and share your styles with users | Requires entering an exact technical prompt, poorly understands natural language |
| Broad capabilities with multiple neural network models under one interface | Issues with rendering faces, hands, and small details |
| Ability to use for free | Requires a paid version for full functionality |
Bing Image Creator
Microsoft has released its search engine with ChatGPT and, consequently, the ability to generate images using DALL-E.
Capabilities and Ease of Use
We previously wrote about DALL-E as the first in our top list, and since Bing Image Creator is a complete analogue of this generator, there is no need to describe everything again. All the information can be found above.
Below are images generated in Bing Image Creator. They are no different from DALL-E.
Jasper Art
Jasper Art is a product of the same company that created Jasper Chat. The first is used when entering requests into the chat like “generate me,” that is, like DALL-E, it is fully integrated with Jasper Chat.
Capabilities and Ease of Use
Unfortunately, we cannot independently examine how Jasper Art works as the free version is available only after entering payment details. In theory, it should interact well based on simple human language, but users on the internet complain that the neural network poorly understands what image a person wants to get from it.
Image Quality
Users on the internet note that the image quality currently leaves much to be desired. Judging by Jasper Chat, most likely the image generator will improve soon.
Performance
Performance is unknown.
Pricing
Pros and Cons
| Pros | Cons |
| An alternative to ChatGPT with DALL-E. It currently does not match the level of image generation, but the generator is constantly evolving. | The neural network does not understand what is expected of it. |
| Free trial for a week. | Requires payment. |
Starry AI
Upon entering the site and registering, we are presented with four options for generations: drawings, photos, illustrations, and creating your own style. Overall, the neural network has a minimalist and very user-friendly interface.
Capabilities and Ease of Use
The neural network prefers precise technical prompts rather than natural human speech, so it is challenging to communicate with it normally. All the typical issues with neural networks are noticeable: ten fingers, repeating elements, and strange eyes and faces.
On the request “generate a homeless person with a laptop and a ginger cat for me,” we got this:
Interestingly, the person is holding an Apple laptop and levitating. But an intriguing moment: there were no issues with the hair color matching the cat or generating a cat’s face instead of a human face. However, in some variations, the cat disappeared altogether.
Image Quality
Most likely, if you delve deeply into the generator, you can achieve good results. However, you can only create up to two images per day, or you’ll have to buy a PRO account.
On a simpler request like “generate an image in the style of Van Gogh’s paintings with a girl in sunglasses,” we got this image:
Performance
Starry AI generates images slower than other generators – around 30 seconds. The computer version significantly lags at all stages of work: issues with registration, starting the process, and downloading finished images.
Pricing
The PRO account costs 9 dollars per month. A mobile application is available for any devices.
Pros and cons
| Pros | Cons |
| Developing, decent quality of generations | Operates slowly |
| Convenient minimalist interface | Requires payment |
Dream Studio (Stable Diffusion)
Dream Studio is an interface for using the Stable Diffusion image generation neural network.
Capabilities and Ease of Use
If Stable Diffusion is a “heavy” neural network that can only be installed on a powerful computer, Dream Studio allows you to use the generator right on its minimalist and user-friendly website with style settings, image format options, and more.
Image Quality
We couldn’t manually verify the image quality due to the difficulty of accessing this neural network. Since this is a user interface and it is in test mode, similar difficulties may arise. However, the gallery of generated images looks impressive.
Speed of Operation
Unknown.
Cost
Initially, the user has access to one hundred credits – approximately 500 free generations. This amount is enough to fully explore the capabilities of the site and decide whether to purchase credits. 1000 credits cost only 10 dollars.
Advantages and Disadvantages
| Pros | Cons |
| Good image quality | Registration issues |
| User-friendly interface | Partially paid |
Conclusion
In the world, there is an ethical struggle between content creators and AI, and companies constantly introduce new rules for using artificial intelligence in their work (especially concerning contest entries, award-winning works, etc.). It is unknown how this ethical war will end, but being able to use generators skillfully and thoughtfully is a useful skill that will soon be required for any relevant job. The key is not to mechanically create images, passing them off as art, but to use AI as a tool to enhance the efficiency of a living person.
Leave a Reply