Welcome to the WeChat subscription number of “Sina Technology”: techsina
Text / Yang Jingxiao
Source: Qubit (ID: QbitAI)
How powerful is AI generation? Humans are already working on it now.
According to a painter netizen, a new type of work has been born in their circle!
Although at first glance, the name is the same as “illustrator”, but take a closer look at the work content: use AI to draw and then filter and refine. Help, this is inappropriate, just follow the AI to do some leftover work~
In other words, you don’t need to have drawing ability at all, as long as you know PS, you can be competent for this position.
As a result, some netizens complained: the job has no prospects but the volume is still large, and the boss is really keeping up with the times.
In addition, the repair work is very complicated, and it is not something ordinary people can do, and the company only gives a salary of 3-4 thousand (part-time).
Can not help but make people question: how is it possible to recruit?
New jobs have been quietly created
Since the recruitment of AI illustrators focuses on “the ability to use AI painting tools”, what is the generation effect of AI tools today?
First of all, for the painting tools themselves, including Midjourney, and various tools based on Stable Diffusion, such as Lexica and KREA, etc., the usage of keyword search and prompt word generation has been opened, which can be used for free.
Taking Lexica as an example, let’s first look at the image effect generated with keywords.
Regardless of Chinese painting or ancient architecture, the color and style of the composition are actually quite good:
Even if the single image is enlarged, the generated eaves details and the arrangement of the buildings in the distance are more attractive:
Looking at another platform KREA, enter the keyword “cyberpunk”, and the style that comes out is also quite modern:
More than just keyword generation. On the Lexica platform, there is a more useful function “Search by Image”.
For example, here we enter a photo of a doctor examining a lung image:
After the AI is detected, the output photos will have similar effects such as “lungs”, “doctors”, and “masks”:
Of course, the effect is not necessarily easy to get started, because in AI painting, the selection of prompt words is actually more critical.
However, if you search based on public information, you will find that many netizens have spontaneously contributed various prompt words with good effects:
The recruiting company also gave a series of AI prompt word combinations that have been tried and worked well:
Finally, there is the issue of the copyright of AI painting.
Although the copyright of many AI painting websites is still unclear, there are already websites like 6pen Art, which clearly state that “copyright is not reserved”, that is, the copyright of all works generated by the author is handed over to the author himself, for personal use and commercial use. or avatars can be:
At the same time, these websites have also given the AI painting models that can be used behind them, but the number of times of use needs to be obtained through recharge and other means.
From the perspective of the company’s recruitment needs, you need to have two skills to be competent for this position:
First, know how to use AI painting software; second, master the ability of PS refinement, and need to organize the work in a clear layer folder.
From this point of view, in order to be competent for this position, there are at least three skills that humans need to master: 1. Learn to generate prompt words by yourself; 2. Test the best AI painting tools; 3. Refine AI to meet Party A’s aesthetic picture.
From the first two points, you don’t even need to master the specific principles of AI painting to complete this work, you just need to learn how to guide it.
“Understand AI painting and accept drafts”
In fact, if it is not limited to the condition of “serious positions”, making money by painting with AI has gradually become one of the default phenomena in the circle.
For example, in some illustration outsourcing trading platforms, some designers have already made money with the help of AI drawing tools.
For example, on the Mi Painter platform, some designers do not shy away from talking about their painting style, and will introduce AI-based painting elements in their works.
In order to emphasize the quality of their paintings and avoid buyers from worrying about copyright issues, painters deliberately marked the words “no AI synthesis” in the introduction. From this, it is not uncommon for painters to generate paintings with AI.
At the same time, there have been many tutorials about AI painting on video websites such as station B.
Some netizens even said that there have been promotional advertisements for teaching AI painting (but it is not clear whether this netizen saw an artificial intelligence painting tool or Adobe illustrator):
For the emergence of artificial intelligence painting-related jobs, many netizens expressed their concerns:
On the one hand, the emergence of this type of work will inevitably lead to further weakening of Party B’s right to speak; on the other hand, the current lack of clarity in the laws related to AI painting may also bring risks to the AI painting itself.
In this way, Party A can break the salary on the grounds that “the original picture was drawn by AI”.
Isn’t the employee responsible for the accident?
But some netizens said that in any case, the emergence of AI painting is becoming a trend.
For example, not only the type of work of AI painting, but now there are even Chinese groups who use machine translation to eat and voice actors who use AI to generate speech.
There are even netizens who graduated from the Tsinghua Academy of Fine Arts in person, saying that they have switched from original painters to typing codes:
How is the business ecology at home and abroad?
But even with such disputes, it has never been able to stop the process of global commercialization.
AI painting (image generation) is actually just a microcosm of AIGC (AI content generation) panorama. In addition to images, there are also technical scenarios such as text, video, policy generation, and cross-modality.
And from the perspective of the entire industry chain, the previous qubit think tank “AIGC/AI Generated Content Industry Outlook Report” has sorted out the domestic industry.
The upstream includes data supply, splitting and labeling, creator ecology, related open source algorithms, and underlying cooperation tools, etc.; the downstream includes various content creation and distribution platforms, AIGC content detection, content terminal manufacturers, consumer goods manufacturers, etc.
Companies whose core is located in the middle reaches are represented by comprehensive track start-ups such as Xiaoice, which have been valued at more than 1 billion US dollars in two years. Specific products include virtual human, audio generation, visual creation, text creation, virtual social networking, etc. A virtual human business model that uses the logic of “human resources” to make commercial quotations is proposed.
Large factories and research institutions also have corresponding layouts, including Baidu, Tencent AI Lab, Ali, Zhiyuan Research Institute, Microsoft Research Asia, SenseTime, etc.
In the vertical field, take the field of image generation as an example. ZMO is one of the representatives. In 2020, an overseas enterprise headquartered in Hangzhou will be established. The core members are from companies such as Google, Apple, Meta, and other companies. In May this year, it completed an A round of financing of 8 million US dollars.
One of the core products is the image generator imgcreator.ai, which can generate arbitrary image content from text. From the official interface, there are currently three operating modes, free version, basic version and newcomer (Starter) version.
Looking around the world, what we most often perceive is the technical race between OpenAI and Google.
But compared to Google, OpenAI’s business model is clearer. At present, it has begun to provide APIs to the outside world, and is divided into four models to charge externally based on usage.
And in the field of vertical image generation, it seems that a hundred flowers are blooming.
Take the popular StableDIffusion as an example. Stability AI, a start-up company headquartered in the United Kingdom, has a valuation of over US$1 billion in one month after the project was released.
And because StableDIffusion is completely open, many companies have been born on this basis, such as StableDIffusion’s search engine Lexica, which has received $5 million in financing.
The qubit think tank has predicted that in the next 2-3 years, AIGC’s startups and commercial implementation plans will continue to increase. By 2030, the AIGC market size will exceed one trillion RMB.
One More Thing
So, why does AI learn to draw so fast?
Some netizens ridiculed that because you gave him learning materials, he really learned.
△ Knowing @HCIO intercepted from Baidu Post Bar
Reference link:
[1] https://https://ift.tt/e8qboWt
[2] Zhihu @HClO (authorized): https://https://ift.tt/uFklgsO
[3] https://ift.tt/FLbfYis
[4] https://lexica.art/
[5] https://www.krea.ai/ ? q=cyberpunk
[6] https://https://ift.tt/muySXBZ
(Disclaimer: This article only represents the author’s point of view and does not represent the position of Sina.com.)
This article is reproduced from: http://finance.sina.com.cn/tech/csj/2022-10-09/doc-imqqsmrp1948058.shtml
This site is for inclusion only, and the copyright belongs to the original author.