這是一副剛剛獲得了美國科羅拉多州博覽會舉辦的藝術比賽一等獎的作品《太空歌劇院》。而該畫的創作者傑森艾倫自曝該作品是其用AI繪畫 App「Midjourney」創作產生的。
《太空歌劇院》的出現，讓人們再次將目光放在AI與藝術的融合上。伴隨著科技和軟體能力的提升，「人人都可以畫畫」的時代已經來臨了。百度文心一格相關負責人表示，電腦與藝術創作的融合，具有很大的想像空間，AI將對藝術創新起到正向促進作用。
每個普通人都可以透過文字來描述自己的想法，然後透過機器來實現創作，人工智慧在藝術領域的學習與創作能力正以蓬勃之勢不斷刷新我們的認知，一個新的藝術類別也逐漸興起。
「我們正在目睹藝術在眼前消亡。」這是諸多網友在得知《太空歌劇院》的創作過程後最直觀的看法。他們認為Allen使用AI生成影像完全是在作弊，但Allen自己卻不以為然。
他介紹了這幅作品的創作過程耗時一個月，不停地修改輸入關鍵字，給MidJourney提交盡可能精確具體的詞語，創作了100多張作品後，挑選出了三張自己認為最滿意的，用GipaPixel AI將圖片清晰化，再用Photoshop進行多次調整，最後才印在畫布上。
這並不是AI作品第一次在正式的藝術舞台上亮相，2018年10月，藝術品拍賣行佳士得曾以約一萬美元的價格成交了一幅由AI生成的油畫《 Edmond Belamy的肖像》。創作者讓AI模型分析了15,000幅14至20世紀的油畫後，仿出了一幅風格類似的作品。
佳士得對作品的形容是：在藝術與人工智慧之間互動的實驗。意味著，這幅拍賣作品的核心價值並不在畫作本身，人們關注更多的是當時看來前所未有的生產方式。這足以說明，AI技術產生的圖片不是既有圖片的拼湊，而是類似人類經過長期繪畫訓練後的「再創作」。
儘管人工智慧作品充滿了衝擊力和視覺張力，但和幾乎所有其他的深度學習模型一樣，在理解知識、推理、邏輯方面做得都不夠好。一個已經被廣泛注意到的問題，是人工智慧經常產生奇形怪狀的手。
這就要先提到AI繪圖的原理。根據百度文心一格相關負責人介紹，AI從海量圖文對應的數據中學習到了「語言描述」與「藝術畫面」的關聯，以隨機化的起點開始，不斷修正畫作，最終形成在美學上與人類經驗與知識高度一致的成品。
人的一隻手有超過 20 個關節，是人類身體上形狀最豐富的結構之一。在大部分用來訓練的圖片中，手部經常不是最核心的部位，不同角度、不同距離、手不同勢、還有被陰影和其他物體遮擋等現象，就會讓AI在手的表現上出現問題。
從技術角度來看，需要提高的主要是兩個方面，包括生成的可控性和細節描述能力。目前AI對於數量、邏輯、關係、多重圖關聯等問題暫無有效的解決方案。例如要求生成2個蘋果，左邊是紅色，右邊是綠色。雖然這邊的關係並不負責，但是在通常情況下，模型並不能穩定的產生正確的結果。
另一方面，AI對於更複雜的、規律細節的描述能力有待提升。例如一棟居民樓的窗戶應該具有多種形態，開著的、關上的、有人在窗邊看風景、有晾在陽台上的衣服等等……但目前的AI繪畫尚未具備這樣的能力，同理，手部繪畫也是這樣的問題。
人工智慧上一次驚艷網路還是2020年。
這一年，由OpenAI公司開發的GPT-3橫空出世。有開發者為GPT-3 做了圖靈測試，發現GPT-3對答如流，正常得不像機器。而最近又一次驚艷網路的ChatGPT同樣出自這家公司。
Compared with GPT-3, the main improvement of ChatGPT is its memory ability. ChatGPT can store conversation information and continue context to achieve continuous dialogue, which is crucial in conversation scenarios and greatly improves the user experience in conversation interaction mode.
ChatGPT is inseparable from big models, big data, and big computing power. According to Guosheng Securities research report data, the ChatGPT technology base uses the fine-tuned GPT-3.5 series model, which has as many as 175 billion model parameters. The public crawler data set mainly used by OpenAI has a human language data set of more than one trillion words. . GPT-3.5 is trained on Microsoft Azure AI supercomputing infrastructure, and the total computing power consumption is about 3640PF-days (that is, one quadrillion calculations per second, running for 3640 full days).
The key technology to provide a high-quality application experience for AI painting is large models. In March 2021, OpenAI released the artificial intelligence painting product DALL·E, which has been upgraded to DALL·E 2; Internet giants such as Google, Meta, Baidu, and Alibaba have also launched large AI models and corresponding products that support content generation.
Baidu launched Wenxin Yige, an AI art and auxiliary creation platform based on the Wenxin model. It is based on Baidu's continuous technological innovation of Fei Paddle and Wenxin model and released in the aspect of "AI drawing" Products and technology tools. Positioned for people with design needs and creativity, it intelligently generates diversified AI creative pictures based on the Wenxin large model to assist creative design and break creative bottlenecks.
Wenxin Yige is a product innovation realized by the Wensheng diagram system based on the Wenxin large model. On the Wenxin Yige official website, users only need to enter their own creative text and select the desired painting style to quickly obtain the corresponding paintings generated by Yige. Wenxin Yige can produce not only magnificent and gorgeous art paintings, but also creative and hyper-realistic pictures with Chinese characteristics. It supports more than ten painting styles and different formats such as Chinese style, animation, illustration, and oil painting.
Baidu Wenxin Large Model has explored a path to industrialization of large models. Baidu CTO Wang Haifeng pointed out that companies with comprehensive advantages in algorithms, computing power and data can encapsulate the complex process of model production and provide large model services to thousands of industries through low-threshold and high-efficiency production platforms, thereby forming a large-scale Model industrialization path.
In 1839, the camera came out, which had a huge impact on the purely manual painting at that time. The well-known French artist Delaroche also lamented: "Painting is dead!" However, with the popularity of cameras, the concept of photography as a work of art has gradually been accepted by the public.
In the famous "Luddite Movement" in Britain in the 19th century, a large number of British workers broke into factories and destroyed textile machines and other machines because they believed that the machines had taken away their jobs.
But in the end, painting is not dead, new forms of expression have emerged in art, and machines have helped society create more productivity. Economist Keynes once analyzed that the emergence of new technologies will lead to technical unemployment. In order to avoid being eliminated, people always have to improve their ability to apply various technologies.
Judging from current practice, AI painting has brought new creative methods and greatly improved painting efficiency. Whether it is for professional visual content creators such as painters, designers, and artists, or textual content creators such as media and authors, AI painting undoubtedly provides creative ideas and assists in artistic creation, improving both efficiency and quality.
In the future, both technology and applications will develop rapidly. From a technical point of view, in the future, AI can not only draw, but also generate AI original short videos based on text content with one click, and automatically add subtitles, voice, music, etc. For example, at the 2022 Vientiane·Baidu Mobile Ecology Conference, "Star", written, composed and sung by digital personality Du Xiaoxiao launched by Baidu APP, is the first MV fully generated by AIGC. The picture part is composed by Wen Xin Yige based on the lyrics. Automatic generated. In terms of application, AI painting will continue to be integrated into various content creation scenarios. In addition to the common fields of illustrators and art design, it will be increasingly widely used in mobile content production, games, industrial design, AI education and other scenarios in the future.
For outstanding artists, it is impossible for them to be replaced by AI. But in order to be more competitive, ordinary artists may need to learn to use AI for artistic creation, just like today's painters must master digital drawing software. Advanced tools will eventually become necessary production means.
As the creator of "Space Opera" said in response to the controversy: "I will not apologize for this, I did not break any rules, and I won."
以上是儘管是透過AI繪製的，我的畫還是獲得了獎項的詳細內容。更多資訊請關注PHP中文網其他相關文章！
