Michelangelo's paintings actually contain the secrets of AI painting? !

With the popularity of AIGC (artificial intelligence generated content) tools such as ChatGPT and Wenxinyiyan, AI (artificial intelligence) technology has quietly integrated into our daily lives, significantly improving our work efficiency and enriching our life experience, while also stimulating our imagination and creativity. In this wave of AI technology, AI painting technology has become a focus in the AI ​​field with its amazing creative results.

So, what exactly is AI painting? What capabilities does it have, and on what principles and technologies does it operate? Let's enter the mysterious world of AI painting and find out!

AI painting unlocks infinite possibilities

AI is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. It is also an important driving force for a new round of technological and industrial revolutions. Among the wide applications of AI technology, AIGC technology is particularly worthy of attention. Based on advanced machine learning models, this technology has achieved the ability to generate a variety of content such as text, images, videos and music by analyzing and learning massive data sets. This not only demonstrates the innovative potential of AI, but also provides great convenience and inspiration for professionals such as content creators, designers, and engineers.

AI paintings

As an application example of AIGC technology, AI painting has occupied a prominent position in the Internet and digital art world. With the help of platforms such as Midjourney, Stable Diffusion and Wenxinyige, AI painting can help people quickly create a large number of high-quality image works. With the characteristics of low cost, high controllability and high efficiency, it plays an important role in many areas of life such as education and entertainment.

Michelangelo's words actually contain the secret of AI painting

"The statue was already in the stone, I just removed the unnecessary parts."

This statement by Italian artist Michelangelo is describing his creative concept and method as a sculptor, but it also reveals the basic principle of AI painting. The process of AI painting, in essence, starts from an initial image containing a lot of random noise, and gradually removes the "excess" noise through the AI ​​algorithm, and finally "carves" a clear and specific image to meet specific needs. The random noise here refers to a random information element in the input data, like the noise in the picture. It cannot be expressed by a clear mathematical formula, and will produce slight changes each time an image is generated, which is used to increase the diversity and creativity of the model.

To understand this process, we can use the AI ​​painting tool Stable Diffusion to explain it. The name of Stable Diffusion itself implies its working principle, which is the "diffusion" process, which is actually a training process. Take the world-famous painting "Mona Lisa" as an example. If we squint our eyes, the picture will begin to become blurry. This is an analogy of "Forward Diffusion" in AI painting. At this stage, AI analyzes blurred images, learns and understands their morphological characteristics, relies on deep learning to extract feature data from a large number of images, and corresponds to their text labels to build a huge database.

Basic principles of diffusion modeling

When we need to generate a Mona Lisa image with a specific style (such as anime style), the trained neural network retrieves relevant features from its database based on the given prompt word and starts the "reverse diffusion" process, which gradually reduces the noise in the image to clarify the image. In this way, the neural network can gradually transform a noisy image into a clear image that meets user needs based on complex algorithms and huge data sets, just like gradually carving a beautiful statue from a stone.

Easily start your creative journey with AI painting

With the advancement and popularization of technology, the use of AI painting has become simpler and more intuitive. The key to controlling this process is to provide AI with a precise text instruction, namely a prompt word. In order for AI to accurately understand our needs, the prompt word needs to include a description of the image theme, painting style, and image parameters. The more detailed the description, the more helpful it is to assist AI in creating works that meet expectations.

Taking the AI ​​painting tool Midjourney as an example, a typical prompt needs to describe the subject, style, setting, composition, lighting and other elements of the image in detail, and also set the image parameters. For example, you can edit the prompt as follows, "An oil painting of a little boy reading in a room, the little boy is wearing a blue shirt, the background is a messy room, dim and soft light, facing the view, the picture size is 16:9", which can better guide AI to generate images.

Midjourney AI prompt words (top) and image generation interface (bottom)

According to the guidance of the prompt word, AI will generate 4 images as output. The "U" and "V" controls on the interface represent the magnified output and optimized modification options respectively, and the number after each button corresponds to one of the 4 generated images. For example, if the first image meets the requirements, click "U1", AI will enlarge and output the image; if the second image is closer to the requirements but needs further optimization, click "V2", AI will generate 4 images again based on the second image. If this batch of images still does not meet the requirements, the user can instruct AI to regenerate 4 images based on the original prompt word by adjusting the prompt word or clicking the loop button on the right side of the interface. These steps constitute the basic operating process for image generation using AI.

The operation of other AI drawing tools is similar. In Baidu's AI drawing tool Wenxin Yige, users only need to give a simple prompt word, set the aspect ratio, drawing style, drawing mode and other parameters in the property bar on the left, and click "Generate Now" to generate beautiful pictures.

AI painting can become cooler and more fun

With the continuous iteration and evolution of AI painting technology, a series of advanced generation methods and image optimization functions have been introduced, which greatly enriched the ways and means for users to create images. These functions not only improve the efficiency and convenience of image generation, but also give users unprecedented ability to customize and optimize their artworks to more accurately meet their personal creative needs. Let's take Midjourney as an example to see how AI painting can be "played".

From pictures to images

When we want to create a new picture that incorporates some elements of an existing picture, we can use the existing picture as a reference and send it to AI along with the prompt words. In this way, the newly created picture will reflect the characteristics of the reference picture to a certain extent. For example, if we have a photo of a cargo ship sailing on the river and want to reinterpret it in the style of an oil painting, as long as we send this photo and the prompt words of the oil painting style to AI, AI will create a brand new painting in the style of an oil painting.

The original image (left) and the image generated by the image (right)

Image Blending

AI can mix different pictures (up to 4). AI will first analyze the content and characteristics of these pictures, and then organically combine them to create a new work. This process sometimes brings some unexpected creative effects. For example, by combining a photo of a little boy playing football with a photo of a garden, AI can create a new picture of a little boy playing football in the garden. This newly generated image can maintain the original characteristics of the little boy and the garden, and the combination of the two scenes does not feel out of place.

The original image (left) and the generated result after mixing the image (right)

Partial repaint

AI also allows users to refine or modify specific areas of an image. This feature greatly enhances the ability to control image details and provides the possibility to create creative image effects. For example, if you want to add new elements to the face or head of a girl in an image, such as sunglasses, a mask or a helmet, the user only needs to use this function to guide AI to adjust specific areas. In this way, the newly added elements can be harmoniously integrated into the original scene, ensuring the consistency and naturalness of the overall image.

The original image (left) and the partially redrawn image (right)

Keep the characters consistent

In the field of AI painting, there has always been a big problem, that is, it is difficult for AI to maintain the consistency of a single character in multiple pictures, which makes it difficult for us to generate some continuous pictures of the same person. However, in the latest Midjourney update, AI can keep the generated character image consistent with the reference picture in various scenes and action poses based on the character portraits and prompt words we provide. The emergence of this feature allows us to use AI to create comic strips, film and television storyboards, and even character photography works.

Original image (left) and AI-generated continuity image (right)

Nowadays, AI technology has been put into practical use in the fields of film and television, office, and medical treatment. With the support of AI, we can easily complete some tedious work tasks and easily implement some creative ideas into reality. Although the current AI painting technology still faces challenges in terms of controllability, resulting in deviations between the actual output results and expectations, the rapid development of technology indicates that it has great potential. AI painting is gradually becoming a key tool in the field of art and design, providing creative people with the opportunity to explore new areas. With the continuous advancement of technology, we look forward to AI painting bringing a higher level of creative ability and ushering in a new era of collaboration and co-creation between humans and AI!

Text/Jiang Bin, Meng Fanmin Photo/Internet