Couple photo generation (AI generated "Chinese Couple" images have become popular on the i

1. Artificial Intelligence Image Robot

Source: Global Times [Global Times reporter Zhao Juehui] A picture of a "Chinese couple" generated by artificial intelligence (AI) has been widely reposted on social media both domestically and internationally. Its realistic visual effect has made many netizens exclaim: "AI is no inferior to human painters." This work is generated by the latest V5 version of the AI image generation tool Midjournal, This tool is another AI product that has received attention recently after GPT-4.

2. Artificial intelligence image generation

In an interview with Global Times reporters, multiple artificial intelligence and internet industry insiders believe that the Midjournal V5 version represents the cutting-edge level of generative AI, and its generated images have been commercialized in certain fields. Through continued learning and technological iteration, it is expected that its capabilities can be further enhanced.

3. High definition artificial intelligence image materials

However, the above-mentioned individuals also mentioned that Midjournal and generative AI have significant disputes over copyright and other issues in their generated works due to training through public data

4. Artificial intelligence image creativity

How did "Chinese Couple Pictures" make a comeback? Midjournal, developed by a studio in the United States, made its debut in March last year and iterated to the V3 version in August of that year, which began to attract some attention. This year's updated V5 version successfully made Midjournal and her works popular, with the representative work being the "Chinese Couple" image mentioned above.

5. Artificial intelligence image materials

In this image, a young Chinese couple is sitting on the rooftop in a jacket and jeans. The background is a Chinese city in the 1990s. Due to its excellent portrayal of light and shadow, character details, and atmosphere, this image is even considered by some netizens to be a real old photo taken, rather than an AI generated image. According to foreign media reports, Midjournal has been continuously improving in image details since its launch.

6. Artificial intelligence image high-definition

Midjournal officially stated that the V5 version has higher image quality, more diverse output, a wider range of styles, support for seamless textures, wider aspect ratios, better image prompts, and a wider dynamic range. In addition, this version greatly optimizes the previous issue of not being able to accurately generate character hands.

7. Artificial Intelligence Picture Cartoon

A painter who has used Midjournal told Global Times that in previous versions, there were many issues with Midjournal's hand depiction, such as incorrect number of fingers or distorted shape, which was also the main basis for distinguishing whether the image was generated by AI. However, the Midjournal V5 version has made significant improvements in this regard, although unnatural hand situations can still be observed.

One of the methods for artificial intelligence images is to use object detection

Before the V5 version, Midjournal had received a series of applications. Last September, a work named space opera won the gold medal in the Colorado State Fair Art Competition. This work was completed by Midjournal last June. The British magazine The Economist used Midjournal to design a cover to echo the cover story theme of "new frontier of AI".

One of the methods for artificial intelligence film reading is to use object detection

Global Times reporters recently also experienced Midjournal, which is currently mainly used in communication software Discord in the form of dialogue with the "Midjournal bot". By entering prompt words, reporters can have the Midjournal bot generate corresponding images, with many prompt words that can be set, as well as requirements for resolution and realistic style.

The process of artificial intelligence film reading is reflected in

Usually, Midjournal bots generate four images in about one minute, and users can also request a regeneration during use. However, reporters have discovered another issue with Midjournal: users can only make requests through prompt words, but cannot control or modify the image generation process. Image generation is more like a "black box".

Therefore, users need to constantly change or adjust the prompt words to make the generated pictures more in line with the needs. Jason Allen, who created the space opera, said to the foreign media that he had generated more than 900 pictures in total before picking out what was behind the satisfactory works. The technology industry told the Global Times reporter that Midjournal is a big model generative AI.

In 2015, a researcher at Google created DeepDream, the first machine learning application for art creation. Since then, AI image generation has entered the public eye. Li Tao, an AI algorithm engineer at a well-known multinational company, told Global Times that based on diffusion models, Midjournal has created its own closed source model and trained it by capturing a large amount of public data.

At the same time, Midjournal also continues to collect user feedback data and iterate on the model, and finally forms the current achievement diffusion model, which belongs to an unsupervised generative model computer vision processing open source software library. In an interview with foreign media, Malik, the director of OpenCV, said that the diffusion model uses the knowledge obtained from text data to understand the semantics of words together and connect with the real world, This enables AI to generate more complex and diverse images without relying on specific datasets.

Malik believes that diffusion models are a major progress of generative AI, because they do not rely on supervised learning by using the knowledge obtained from supervised learning. These models can generate complex and diverse images without manually marking data, making them more flexible. For some time, many companies have launched AI image generative model or related products.

Recently, Microsoft Bing announced the integration of OpenAI's AI drawing model, DALL · E, and the launch of AI generation function. Adobe has also added an AI drawing product for its Firefly model, "I can no longer believe my eyes." Although Midjournal has gained millions of users with its powerful image generation function, it has also received a lot of criticism and questioning, the first of which is copyright issues.

The website of Forbes magazine in the United States states that, like all AI image generation tools, Midjournal has collected a large number of images from the internet for training, enabling it to create seemingly new art works, which has sparked dissatisfaction among human creators. Currently, many institutions and individuals have filed lawsuits against Midjournal, accusing the company of widespread copyright infringement.

In addition, according to media reports, the US Copyright Office recently announced that AI images generated by Midjournal and other platforms are not protected by copyright. In the description document, the US Copyright Office believes that works automatically generated by Midjournal and other platforms are automatically completed by robots throughout the creation process, and the training data is based on works created by humans, so they are not protected by copyright law.

Xia Hailong, a lawyer at Shanghai Shenlun Law Firm, told Global Times that AI generation tools have brought a huge impact on human copyright law. On the one hand, compared to the massive AI generated works, even if there are infringement cases, it is difficult for rights holders to detect them in a timely manner; On the other hand, in the era of AIGC (Artificial Intelligence Generated Content), lengthy and expensive judicial procedures have led to increasingly inadequate copyright protection.

Human copyright laws and copyright protection agencies may have to make significant adjustments. In addition to copyright issues, the increasingly realistic AI generated content also raises concerns that these images and videos may be used for illegal purposes, such as fraud using AI generated images. Former US President Trump recently announced that he will be arrested on the 21st.

Subsequently, a large number of images of "Trump arrested" generated through AI were spread on social media, misleading some netizens in India. News18 News Network mentioned that recently, some netizens generated a picture of former US President Obama and former German Chancellor Merkel eating ice cream together by the sea through AI, which made people even more worried about what kind of deep fraud would occur in the future.

The website cited comments from netizens stating that "this technology is becoming increasingly terrifying and sophisticated. Currently, AI can create almost perfect celebrity voices or images, and we may soon be unable to trust our ears and eyes." In addition, some parents have expressed concerns to Global Times reporters about the impact of AI generating inappropriate images on children.

Internet expert Guo Tao told Global Times reporters that the current AIGC technology is still in the early exploration stage, and relevant laws and regulations, standard systems, and ethical norms are not yet sound. There are issues such as data security risks, copyright infringement risks, deep forgery risks, trade secret leakage risks, and illegal information dissemination risks.

Relevant departments need to establish and improve relevant laws, regulations, and standard systems as soon as possible, strengthen content review and filtering, and promote the healthy and sustainable development of the industry