The most comprehensive comparison: Which is the best generative AI? A large collection of advantages and disadvantages of the three top image generation tools, you will understand after reading it

Text: Tri Editor: VickyXiao

Image source: Generated by Unbounded AI tool

The hottest keyword in 2023 is probably AI.

From ChatGPT, which was born out of the blue, to Midjourney, which is popular all over the Internet, and then to the global AI large-scale melee, in less than a year, AIGC has continued to explode, stirring up the world.

No matter at home or abroad, everyone seems to have found the ticket to the "new world", either in research and development of AI, or in business + AI. It can be said that AI has become the most certain certainty at present, leading the public to the next AIGC era and realizing a new productivity leap. There are many AI-related tools at home and abroad, but they are all extended from the three mainstream algorithms of Midjourney (MJ), Stable Diffusion (SD) and DALL-E. DALL-E has been upgraded to DALL-E2.

Today we will explore together, what are the advantages, disadvantages and differences of these three mainstream algorithms?

Background introduction of the three "top streams"

Midjourney is a software currently installed on the Discord platform. It was developed by a studio in the United States and debuted in March 2022. In just one year, MJ updated and iterated to the V5 version, and quickly became popular with its artistry. . The generated maps of these explosions on social networks are all generated by MJ. With its amazing rendering quality and perfect commercialization model, it has successfully attracted a large number of paying users and achieved annual revenue of 100 million US dollars.

Stable Diffusion is developed by the British startup company Stability AI in cooperation with several academic researchers and non-profit organizations. It is free to use. The current source code and model of SD have been open sourced and can be used locally on Mac, PC or even mobile devices. Deployment (requires relatively high equipment configuration), you can DIY model, many software tools at home and abroad, most of them are developed with SD as the underlying logic, and its controllability and rendering capabilities are also very strong, and now many designers use it to carry out Some side work.

DALL-E 2 was developed and launched by the OpenAI team. Unlike Midjourney, the pictures generated under simple instructions generally tend to be realistic. DALL-E 2 is the best of the three in terms of realistic images and concise prompts.

Expressive Ability

Let's take a look at their expressive ability together. Under the description of same keyword, how is their image feedback?

DALL·E 2 (left), Midjourney (middle) and Stable Diffusion (right)

DALL·E 2 (left), Midjourney (middle) and Stable Diffusion (right)

DALL·E 2 (left), Midjourney (middle) and Stable Diffusion (right)

From the comparison in the above figure, we can see that DALL-E 2 can handle images of nature and people very well. At present, by using the API of DALL-E 2, three methods of interacting with images can be realized: According to the text prompt Create images from scratch, create edits of existing images based on new text prompts, create variations of existing images.

Midjourney has rich colors and realistic graphics in all attempts. It is capable of generating images based on user prompts. Midjourney is good at adapting the actual art style to create images with any combination of effects the user desires. It excels at environmental effects, especially fantasy and sci-fi scenes that look like game art.

And Stable Diffusion is an open source model that everyone can use. It has a relatively good understanding of contemporary art images and can produce artwork full of details. It's just that it is difficult for ordinary users to use. At present, the most popular discussion among everyone is who is better, SD or MJ, so we tested some images.

Midjourney (left) and Stable Diffusion (right)

Midjourney (left) and Stable Diffusion (right)

From the details of cartoon character image generation, SD is better. With the same keyword generation, the facial features generated by SD are more three-dimensional and exquisite, and Midjourney is more artistic.

Midjourney (left) and Stable Diffusion (right)

When generating realistic illustration-style images, the two are equally effective. The biggest advantage of SD is that users can remix through model embedding, LoRA or network, which can produce unexpected effects.

Midjourney (left) and Stable Diffusion (right)

Stable Diffusion currently has over a thousand models available for download. Each model can be further modified using LoRA models, embedding models, and hypernetworks; Midjourney is limited in terms of models. Only v1 to v5 models are available, and some special models such as niji, test, testp and HD. There is an additional parameter to "stylize" the image. But overall it still pales in comparison to Stable Diffusion.

user's guidance

In fact, these AI image generators have their own advantages. Based on the experience of these three algorithms, we have summarized some salient points and differences. The specific choice can be determined according to your own needs.

1、DALL-E2

The interface of this model is simple and easy to understand, and you can easily generate images without registering a third-party platform. You only need to visit the website and enter your keyword text into the generation box to create images.

advantage:

l Easy to use

l Creativity with flexibility

l No third-party platform required

shortcoming:

l simple image

l The accuracy is not high

l Can only create square images

2、Midjourney

Unlike the DALL-E 2, you need to create a Discord account first, and you have to enter commands like text or pictures to generate images.

advantage:

l High quality image

l Users can customize the image size ratio

l Flexible control of image parameters

shortcoming:

l It is more complicated to use

l Require users to register on Discord

l After the fast time runs out, the image generation time will increase dramatically

3、Stable Diffusion

Both Stable Diffusion and DALL·E-2 are based on the diffusion model, which can draw images according to the input text descriptors (). To complete the production of excellent images on Stable Diffusion requires the correct model + accurate prompt words + parameter adjustment + post-processing technology.

advantage:

l Web operation and installation of open source models to other platforms

l More creative freedom

l A large number of controls for customizing image parameters

shortcoming:

l Requires professional and powerful hardware equipment

l The image quality is very unstable, it depends on the version you use

l High learning difficulty, difficult for ordinary users to control

To sum up, if you have a clear target output, for example, you are an architectural designer, then Stable Diffusion will better meet your creative needs because it is more controllable. And if you don't have a clear creative control goal and want to do some creative work with divergent thinking, then the convenience and low threshold of Midjourney and DALL-E 2 will be your best choice.

Which AI tool do you prefer to use? How is it applied in work life? Welcome to share with us in the comments!