What is AIGC: Understand AIGC in one article
What is AIGC?
AIGC is the abbreviation of English AI-generated content, which is artificial intelligence-generated content.
Simply put, it uses some prompt words to generate text content, pictures, videos, animations, and even code.
The principle of AIGC
The principle of AIGC is actually very simple. With the development of artificial intelligence up to now, the ability of "emergence" has appeared. AIGC uses the "emergence" ability of artificial intelligence to allow artificial intelligence to imagine, reason, and Analyze, and ultimately, “emerge” content.
The meaning of AIGC
AIGC will change the entire content industry.
AIGC has greatly enhanced the productivity of text content, picture content, video content, and animation content, and improved content production efficiency. Therefore, in the future, a large amount of high-quality content will be produced or assisted in production through artificial intelligence.
History of AIGC
AIGC, starting in 2022.
The release of GPT3 and ChatGPT at the end of 2022 will mark the advent of a new era of content production. The concept of AIGC will also become popular in 2023.
Around the world, a large number of AIGC entrepreneurial teams and individuals have started a new round of entrepreneurial boom.
According to data from Google, search volume related to AIGC has increased sharply.
The foundation of AIGC—Large Language Model (LLM)
Large Language Model (LLM), referred to as large model, is the basis of AIGC. The emergent ability of artificial intelligence comes from large language models.
So far, large models can be divided into the following categories according to different uses: text, picture, and video.
1. Text class
This is the basic large model. Mainly used to implement dialogue (chatbot), text generation, code generation, etc. The most famous ones are the following:
GPT
For conversational and textual content generation.
The large model created by OPenAI is the most commonly used large model now. ChatGPT is based on the GPT large model. Closed source large model.
PLM
For conversational and textual content generation.
The large model created by Google is currently a large model that is relatively close to GPT in all aspects. Whether it will be open source in the future is uncertain.
Llama
For conversational and textual content generation.
A large model created by MATE. Open source large models.
It should be one of the best open source large-scale models for conversation and text content generation.
Oncidium
For conversational and textual content generation.
Large model created by Baidu. It is currently the best large model that supports Chinese. Closed source.
2. Picture category
The large image model is a large model used to generate images. It can generate text to image (ie: text to image), and image to generate image (ie: image to image). Currently, the most popular large picture models are:
Midjourney
Used for image generation.
Created by a small team, it is now the most popular large model for image generation. Closed source.
Stable Diffusion
Used for image generation.
Built by a small team. Close to Midjourney. Open source.
DALL.3
Used for image generation.
Built by OpenAI. Closed source.
3. Video large model
The video class large model is a large model used to generate videos. There are three main types:
Text to video generation: text to video, directly generates video through prompt words (text).
Image to video: image to video, generate video through pictures, or pictures + text.
Video to video: video to video, generate video through video, or video + text.
Currently, the most popular large video models include the following:
Stable Video Diffusion
Created by stable.ai, it is the most popular large-scale video generation model and is open source.
Gen-2
Made by runway. It is currently the most mature large-scale video generation model. Closed source.
A major feature of Gen-2 is its emphasis on realism. Therefore, videos generated using Gen-2 are often extremely realistic.
The key to AIGC - prompt, context and AI agent
1. Prompt word (prompt)
The way to interact with large models is to use prompts.
Chat, generate text, generate images, generate videos. . . They are all implemented through interaction between prompt words and large models.
What kind of content is generated? What is the quality of the content? Under the same premise of the large model, different prompt words will make the final results vary greatly.
The so-called interaction with artificial intelligence using natural language refers to prompt words.
2. Context
Large models can understand prompt words based on a certain length of context. Each large model also has different limits on context length. Generally speaking, the longer the context length, the closer the understanding of the prompt word is to the idea of the person who proposed the prompt word.
For AIGC applications, there are multi-round dialogue mode and single-round dialogue mode.
Multi-turn dialogue mode: using large model context.
Single-turn dialogue mode: no large model context is used.
In fact, for many content generation scenarios, no context is needed, only a single-round dialogue mode is enough.
3. AI Agent
An AI agent, simply put, is a role defined for a large model. First define a role for the large model, and then let this role complete the tasks you specify.
You may be wondering, why do you do this?
In fact, it is easy to understand. For example, let GPT write a short story. If you write a short story with the same theme and the same title, if you do not define the characters, and if GPT defines the characters as "Tolstoy" and "J.K. Rowling" respectively, three Comparing the results of these situations, the difference is huge.
you can try.
AIGC application direction
1. Conversational applications (Chatbot)
Question and answer, AI girlfriend, role-playing, etc. are the main directions of current dialogue applications.
2. Text content generation (Text Generate)
Novel generation, article generation, paper generation, marketing copy generation, etc.
Text content generation is currently the most popular application in the direction of AIGC.
3. Image content generation (AI Image)
4. Video generation (AI Video)
5. Animation generation (AI Anime)
6. Code generation (AI Code)
AIGC industry pattern
1. Large model manufacturers
The core of the AIGC era is large models. Continuously improving the capabilities of large models will be the direction in which large model manufacturers continue to strive. Each large model will form an application ecosystem. The profits of large model manufacturers can be either by providing direct application profits to users or by charging fees to developers in their own ecosystem.
2. AIGC application manufacturers
On the one hand, it is oriented to the needs of users, and on the other hand, it is oriented to the capabilities of large models. Find the most suitable application points, growth points and profit points in two aspects.
3. AIGC application distribution manufacturer
With the emergence of a large number of AIGC applications, the demand for the distribution of AIGC applications will also increase. More AIGC applications are built on the web, so the openness of the web has led to the emergence of a large number of distribution vendors. This is more like a navigation station in the PC Internet era, rather than a relatively closed application store in the mobile Internet era.
How to start your AIGC business?
1. If you want to use AIGC applications to improve your productivity
You need an AIGC navigation station that is updated frequently. Or follow oftenProduct HunterAIGC on the website, because many AIGC applications will be published on this website as soon as possible after they are launched to gain more attention.
2. If you want to make AIGC applications
First of all, you need to be familiar with the capabilities of each large model. Play more with GPT and other large models.
Then, you need to find a good application scenario and application point.
Do these two things well and make an AIGC application. It is a very simple matter.
