The Best AI Art Generator?

The Best AI Art Generator?

Stable Diffusion vs Midjourney vs DALL-E 2 vs Leonardo.ai vs niji-journey

In recent years, AI technology has advanced rapidly, and besides ChatGPT, AI image generation technology has received great attention. The most well-known ones are probably Stable Diffusion and Midjourney. In addition, there is DALL-E 2 developed by OpenAI, the parent company of ChatGPT, which specializes in generating images for game development, Leonardo.ai, and niji-journey developed jointly by Japanese company Spellbrush and Midjourney technology, which have all attracted a lot of attention.

Next, let me introduce each AI and review them:


DALL-E 2

ChatGPT’s sibling for generating images

Sample image from DALL-E 2

Sample image from DALL-E 2

Firstly, there’s DALL-E 2, which was announced in April 2022 and developed by OpenAI, the parent company of ChatGPT that we are familiar with. As the second generation of DALL-E, its performance has made significant progress, and the generated image of a spaceman riding a horse has become a milestone in AI image generation, making the public aware of the possibilities of AI-generated images.

However, as an “early” AI Art generator, DALL-E 2’s performance is not outstanding, and currently, it requires purchasing credits to use (there used to be free credits). It may need breakthrough performance improvements in DALL-E 3 to re-enter the discussion of the best AI Art generation.

Review of DALL-E 2:

  • The first AI Art generator I came across, and I found it extremely novel and interesting.

  • You can log in and use it with an OpenAI account.

  • It’s easy to use, simply by using prompts, but there aren’t too many settings.

  • You can choose between text-to-image and image-to-image.

  • It’s a bit expensive at USD$15 / 115 credits.

  • Like ChatGPT, it provides a paid API, which is convenient for development.

Price of DALL-E 2’s API

Looking forward to DALL-E 3 having breakthrough technological improvements to compete for the best AI Art generator again!


Midjourney V5

Generate high-quality images just by playing around on Discord.

Sample Image from Midjourney

Sample Image from Midjourney

Sample Image from Midjourney

Sample Image from Midjourney

Midjourney made its debut in April 2022 and quickly rose to version 5 by March 2023, attracting attention from all over the world with its powerful image-generation capabilities. Its ability to generate realistic human images with extreme precision, and transform them into different art styles has impressed many.

With rich imagination and appropriate prompts, you can almost generate any image you want, including surreal ones, and the generated images are very high quality. It can be said that Midjourney is currently the best AI Art generator.

To use Midjourney, users must join the Discord server and “compete for computing power” with other users. However, this process also allows users to see other people’s prompts for reference, and the generated images can be fine-tuned and enlarged. However, generating images of the same series or character is challenging due to the high degree of randomness. Even when using the same prompt, there will be different outputs.

https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html

Recently, a new feature was added that allows users to upload images to describe prompts, making it easier to output desired effects through trial and error. It is said that version 6 will be released in a few months, and we are excited to see what breakthroughs it will bring.

Review of Midjourney :

  • Extremely powerful image generation capability with exquisite results.

  • Requires logging in through Discord and inputting a prompt in the chat box to use.

  • Detailed instructions on using prompts are available in the official documentation.

  • Practical functions include the ability to choose from four images and fine-tune or upscale after generating.

  • After 25 trial uses, a subscription is required to continue using the service.

  • The generated images are of extremely high quality, making them the top AI Art generator.

  • The large randomness in the generated results makes it difficult to consistently output the same image.

The most powerful AI image generation tool that requires no experience to use!

Homepage of Midjourney: https://www.midjourney.com/home/


Stable Diffusion

Open-source AI art generator

Image from Stable Diffusion

Image from Stable Diffusion

In August 2022, Stability AI released Stable Diffusion, which is different from the AI art generator mentioned above as it is released as an open source and can be tested on Stable Diffusion Online. However, Stable Diffusion Online only provides a Playground similar to DALL-E 2, which only allows the use of Prompt, and does not fully unleash the potential of Stable Diffusion.

In the following months, thanks to the benefits of being Open Source, a lot of developers started to develop apps for Stable Diffusion, such as Stable Diffusion WebUI, Stable Diffusion UI, Draw Things, and more.

These apps can be installed locally and rely on the computational power of your personal computer’s CPU/GPU to generate images, making it possible to generate unlimited images for free, with full copyright ownership. In addition to using the local computational power of your computer, you can also use cloud computing like Google Colab to generate images, and the speed can be increased according to your budget!

Users can choose different platforms for image generation. In addition to the basic text-to-image (text2img) functionality, there are negative prompts available to prevent unwanted content from appearing. Solutions for weaknesses in AI-generated images continue to emerge, such as depth-to-image (depth2img) technology based on image-to-image (img2img) techniques, ControlNet for controlling human poses, and Openpose and DepthLib for addressing body posture and hand deformation issues.

The problem of lack of control over the poses generated by AI has been solved with ControlNet and Openpose.

Later on, MultiDiffusion was introduced to solve the problem of a single Prompt being unable to control the simultaneous appearance of multiple entities. It enables the natural integration of multiple objects such as foreground and background, using specified Prompts to accurately describe them and combine them into a single image.

MultiDiffusion allows users to control the generation area using colored blocks.

In addition to continuous functional updates, due to the ability for users to train their models, a large number of checkpoints and LoRAs with different themes and styles have appeared, allowing Stable Diffusion users to use different combinations of models according to their specific requirements to generate their desired effects.

Download Models: https://civitai.com/

Recently, Stability AI announced the beta release of Stable Diffusion XL, which utilizes a more powerful model for training and demonstrates impressive capabilities.

Image from Stable Diffusion XL

Image from Stable Diffusion XL

If you are interested, you can try it out for free on Stability AI’s online AI creative platform, DreamStudio. In addition, DreamStudio will continue to introduce many powerful features, making it a strong competitor to Midjourney.

Review of Stable Diffusion:

  • Stable Diffusion WebUI requires some Python knowledge and may be difficult for beginners without experience.

  • Different styles require downloading different models, which can take up a lot of space (from 500GB to several TB for multiple models) and time (especially for checkpoints of 5 to 8GB).

  • High-end Nvidia graphics cards are required for PCs, while Macs need to have the level of M1/M2 Max, and at least 32GB of memory is recommended for computing high-resolution images.

  • Despite the relatively simple interface of Stable Diffusion WebUI, the learning curve is steep due to the many settings and features, and novice users may experience crashes or produce strange images due to unfamiliarity with prompts.

  • The open-source community is growing rapidly, and the development of new features is constantly evolving. Being able to train models personally is a unique and exciting prospect with great growth potential.

  • Once familiar with the settings, it is possible to 100% reproduce the same image, and with more and more companies using Stable Diffusion as a computing foundation, it is expected to become the most popular AI art generator. With the abundance of open-source resources available, the potential for rapid development is high.

An AI tool with rapid growth has the potential to become the best when used in the hands of experts.


Leonardo.AI

Integrated creative image generation tool combining Stable Diffusion and Civitai.com.

Images from Leonardo.AI

Images from Leonardo.AI

Images from Leonardo.AI

Many Midjourney and Stable Diffusion users have expressed their frustration with the limited options and difficulty of use. However, Leonardo.AI has emerged as a solution for many users.

Built on Stable Diffusion technology, Leonardo.AI offers a more user-friendly UI than Stable Diffusion WebUI and integrates with a community similar to Civitai. Users simply select a desired image, apply the desired style, and use AI-suggested prompts to easily generate the desired image. While Leonardo.AI is primarily marketed as a tool for game graphics, it can also be used to create realistic human images by remixing appropriate models.

Furthermore, since Leonardo.AI operates in the cloud, it is likely to be faster than your personal computer. Users receive 150 free tokens per day and are subject to some restrictions, with 5–8 tokens required to generate each image, but basic usage is free. Additionally, Leonardo.AI includes two cloud-based free model training capabilities per month, making it an incredibly generous tool.

You can also subscribe to their plan if you need to use it in large quantities.

Review of Leonardo.AI :

  • Using it after being frustrated with Stable Diffusion WebUI was particularly touching

  • Fast generation speed, simple interface and powerful features

  • Can directly apply styles, saving time on finding and installing models

  • The pre-trained models that are already available have been optimized to a certain extent, and have much less chance of crashing than Stable Diffusion

  • Due to being based on Stable Diffusion, it is expected to grow rapidly in the future

  • 2 Free Models Training usage per month

  • It can be used for free as long as it is not used excessively (if they don’t change their policy)

The most powerful choice that balances difficulty and control level, while also being free, simple, and easy to use!


niji-journey

Midjourney for Anime from Japan

Image from niji-journey

niji-journey, developed by Spellbrush in collaboration with Midjourney, is the hottest AI image generation tool in Japan that specializes in generating anime-style images. It operates in the same way as Midjourney, with the only difference being that niji-journey focuses solely on generating anime-style images. The technology behind niji-journey is 100% provided by Midjourney, and the pricing model is the same as well.

Thanks to the upgrade to Midjourney v5, niji-journey v5 produces higher quality images with more varied styles and finer lines. However, with the addition of the -nijicommand in the Prompt, Midjourney v5 can now simulate niji-journey style, causing some confusion for thousand of Midjourney users.

One of the biggest advantages of niji-journey is that it supports four languages: Chinese, English, Japanese, and Korean, allowing users to generate images using their preferred language. This is especially helpful for users who struggle with English vocabulary and can generate images using different language prompts.

In summary, niji-journey v5 offers the same powerful and user-friendly experience as Midjourney, with the bonus of generating anime-style images and supporting multiple languages. The upgraded image quality with finer lines is a big plus for users who prefer detailed anime images. Finally, the pricing is as same as Midjourney.

The most powerful AI art generator for anime!

Homepage of NijJourney: https://nijijourney.com/en/


Summary

Except for DALL-E 2 which is currently falling behind, each AI art generator has its strengths and is the strongest in its respective field. However, in terms of overall capabilities, I prefer Leonardo. AI. It has the ability of Stable Diffusion and the benefits of cloud software and is also quite generous in providing a considerable number of free Tokens every day. As long as its pricing policy does not change, it should have considerable competitiveness.

I am looking forward to the next AI drawing tool that will emerge and may have the ability to output 3D models directly!