When thinking of the top trends of 2023, I believe AI must be on the list.
From text to pictures, and even video, AI can quickly generate enough content to confuse the real with the unreal, arousing people's interest and generating discussions across the internet.
If we look at the constantly shifting opinions on the Internet, we can clearly see that people's mentality has gradually changed from curiosity and play toward questioning the authenticity of AI-generated content, and worries about an AI takeover. In fact, the development of science and technology will inevitably bring about changes in the way we work. Just as cars replaced carriages in the past, the demand for carriage drivers has declined, but the demand for car drivers has increased.
Let's take AI art as an example. If we search for related topics on social media, we can see how netizens are already discussing that some companies require their creatives to learn how to use AI art platforms. Some even said that in their companies, if they failed to learn it before the given deadline, they might lose their job.
This is similar to the case of cars replacing carriages. In the past, the tools for painting were pen and paper, then software such as Photoshop and Illustrator came in, and in the future (or we should say "now") these tools may be replaced by AI art platforms. What does not change is the original creativity (beginning) and the quality control (ending), and what changes is the tool and the intermediate process. As content producers, we need to continue mastering these two essential abilities which are difficult to replace by AI - original creativity at the "beginning" and quality control at the "end." When technologies change in the intermediate process, we need to master new tools quickly.
1. What is AI art?
AI art is a form of painting realized by AI (artificial intelligence), which uses algorithms and machine learning models to generate works of art. These techniques make it possible for computers to simulate the painting skills and styles of human artists and produce visually satisfying results. The application of AI art includes image processing, virtual reality, game development, animation production, and other fields.
2. What is a prompt?
In AI painting, prompt refers to the text or image input by a user, which is used to guide the algorithm to generate artwork. Prompts usually include keywords and phrases that describe the subject, style, and visual elements of the work, and the algorithm generates a new work of art based on these inputs. A picture can also be a prompt, and the algorithm tries to build on that picture. In fact, we could define the prompt as the instruction of "guiding" AI to generate the required content by inputting text and images.
In the words of advertisers, it is equivalent to a "brief" to AI. Since AI is only a tool, it does not know what we want, and it is often difficult to generate the desired results in one step, so it needs to be guided and fine-tuned many times with prompts until the desired results are generated. Netizens in China vividly gave the prompt the nickname of "chanting". Just like a magician in the magic world, it conjures up the desired results by chanting incantations.
Now there is even a new profession called Prompt Engineer. According to information shared on the Internet, in December 2022, the first officially hired prompt engineer made their appearance. AI painting and crafting prompts may soon become basic skills in the creative industries, just like Photoshop skills, which have long been written into job requirements.
3. Who might benefit from AI painting and how?
Artists and designers
Creating sketches, prototypes, and design concepts more quickly, at least as inspiration or drafts, which can be further processed by hand to improve efficiency and save time.
Game developers
Generating the initial drafts of game elements such as environments, characters, and objects, to save production costs and time, and improve the visual effects of the game.
Media and advertising industries
Assisting in producing visual elements in media and advertising, such as posters, billboards, logos, and cartoons.
Architects and urban planners
Finding inspiration for building exteriors and interior layouts, as well as the visual presentation of urban planning schemes.
Industrial designers
Generating drafts of the appearance design and models of products, and further improving the efficiency and quality of product development.
4. How to get started with AI art?
As we say in China, to know the taste of pears, we must taste them ourselves. Taking Stable Diffusion (SD), one of the most popular AI art platforms, as an example, we refer to the following five steps:
1) Register or install an AI art platform
To use the platform, we can directly register and operate it on the Stable Diffusion official website (note that a free account will have usage limits), or we can install and deploy it locally on our own computer.
2) Start learning from imitation of existing prompts
Just as with learning a new language, we can start with imitation. Learning prompts is not difficult (compared to other programming languages), but there are certain rules and thresholds. If we formulate prompts from scratch according to everyday spoken language, it is difficult to obtain desirable results.
A quick way to get started is to copy a prompt, that is to say, learn by imitation. In some AI art communities, netizens often share their own works and corresponding prompts, including input instructions and settings. We can first copy and paste the prompt of some works to imitate and generate similar works, and then fine-tune specific prompts to get the results we want.
3) Learn the "grammar" and "vocabulary" of prompts
"Grammar" includes rules and writing methods in the prompts, just like in written language. For example, commas are used to separate different phrases, and brackets are used to emphasise or enhance weight. "Vocabulary" refers to some common phrases in the prompts. When entering text in the prompt box, some suggested phrases will automatically pop up in the drop-down menu for assistance, and we can use the Prompt Search Engine of Stable Diffusion's official website as a supplement to understand the actual writing of prompts "phrase collocation" as a reference. Once we have learned the "grammar" and "vocabulary", we can gradually stop imitating prompts and start talking with AI from scratch to generate the desired content.
After doing the above three steps, we can at least start generating content. However, we may find that the generated works are still different from those corresponding to online prompts and the result may not be what we expect. Why is that? Because the works on the Internet may use different models. If we want to imitate the image better, we need to download the corresponding model files.
4) Download more customised models
There are AI models in the installation package, which can be used directly. If we want to experience different effects and styles, we need to download more customized models.
The downloaded customized master model files are placed in the Models/stable-diffusion folder under the local root directory.
5) Download fine-tuning models
There are many types of fine-tuning models, such as DreamBooth, TextualInversion, ControlNet, LoRA, and others. Take LoRA as an example, which literally translates as "Low-Rank Adaptation of Large Language Models". It can be understood as a small model or plug-in based on a large model, which has trained the large model in a certain direction and "frozen" the relevant parameters so that users no longer need to train the model in a certain direction from scratch. Users can then directly add or modify them according to their own tastes.
For example, if we have the LoRA of a game character, we can add it to the prompt to include the character's features in the image. Of course, if we can't find a ready-made fine-tuning model, we can also train our own model.
The downloaded LoRA file is placed in the Models/LoRA folder in the local root directory.
Through the above five steps, we should be able to start to experience AI art without the need for advanced computer skills. Of course, it is easy to get started with but can be difficult to master. To become proficient in AI painting, we need a lot of practice. I am also still learning, and I believe that with the further evolution and wider application of AI (such as Microsoft's integration of AI into the Office suite), the threshold for AI-generated content will be lower and lower in the future.
(The Chinese version of this article was first published on Forbes China)
©️All photos designed by Freepik
About the author
Leon is a seasoned veteran in digital marketing communications with over 19 years of experience. He held digital-related positions in various global agencies and organizations, such as BCW/Burson-Marsteller, MediaCom/GroupM, MRM/McCann, and Microsoft. Outside his professional career, he actively engages with the digital community. Leon is a columnist for Forbes China and a professional lecturer at the Communication University of China.
- name
- Leon Zhang
- role
- Head of Digital Consultancy
- location
- Beijing, China