This tutorial applies to any industry and has universal applicability. All content is shared based on the moderators’ personal experience and verified results; it also covers the knowledge points and tools we commonly use in our daily work. To make it easier for everyone to use, we have also organized it for you.
[baike]Chatbots[/baike]
Chatbots are the foundation of getting started with AI and are also essential. Popular ones include ChatGPT, Google’s Gemini, Baidu’s ERNIE Bot, xAI, etc. Their content output is based on the collection of globally known content, including websites, books, purchased data, etc. Once a bot has learned this content, it provides us with the following main capabilities. Based on these capabilities, bots can do work for many industries, and in some fields they even surpass humans, because they do not get tired and rarely make mistakes; they have almost none of the weaknesses of human nature. Data is the foundation that supports any form of content, so a chatbot is your first tool. To accommodate beginners, we explain each item one by one to ensure everyone can understand it and gain something from it.
- Knowledge Q&A
- Content Generation
- Content Training
- Analysis and Screening
1. Knowledge Q&A
Use human language—your native language—to converse with and ask the bot questions. It will tell you everything it knows; it is almost omniscient and omnipotent, and its effect is especially obvious with cross-domain tasks.
Examples:
1) Ask it how to use AI, what AI includes, what it can and cannot do.
2) If you want to make comics, you can ask it about the comic production process, implementation methods, and implementation process.
3) If you want to build a website, state your goal, and it will start writing code for you and tell you how to set up the website.
4) If you have some resources and want to monetize them or need some channels, you can discuss models and tactics with it and what kind of cooperation you should look for.
In the process of asking questions and getting answers, you will understand the bot’s absorption and feedback logic, and you will gradually learn how to communicate with it to get the content you want. This is prompt engineering—how to phrase things more skillfully and how to quickly get the answers you want.
2. Content Generation
Content generation is the most prominent capability of this generation of AI, including an upgraded version of Q&A. It can organize language to give you the articles, papers, novels, materials, tables, etc. you want—almost any form based on text. During content generation, you will understand what kind of content the bot generates, its writing style, writing structure, writing techniques, etc. At first, it may not meet your expectations; this requires a learning process.
Its content output is random, so when many people ask the same question, its answers will almost never be identical. Precisely because of this randomness, your answers will vary greatly. For example, you can ask it to write an article of 1,000 words, divided into several paragraphs, with what to write in the first paragraph, the second paragraph, and the third paragraph, and whether to write in a straightforward, down-to-earth style or in the style of a writer/editor—all of this is defined by you.
From what we know so far, some people who are good at using ChatGPT can make good use of it to get the content they want. The content looks sophisticated and is hard to identify as bot-generated. How to efficiently produce high-quality content requires strong prompt skills and, even more, a good idea.
3. Content Training
We mentioned earlier that its knowledge source is collection, which means its knowledge base is constantly updated. It continuously acquires the latest free content and purchases targeted data. We call this contemporary AI. You can imagine that what it knows is not necessarily enough. We all know that quite a lot of valuable content is confidential or paid, so in some fields it cannot output high-quality content. But at the same time, you want to use it specifically to produce this high-quality content—then training and targeted output are involved.
Methods and steps 1. Collect content 2. Upload content 3. Output instructions
When you do not have private data, you must learn to collect it. For example, if you want all your articles to be like those of a certain author on Zhihu or a WeChat official account, and to generate articles in that style, then you need to collect content. The simplest method is copy and paste. Copy some content into a txt document; this is your knowledge base. Then upload the knowledge base to the bot and let it learn, then tell it to write an article titled XX and refer to the writing style of the TXT knowledge base. You will find that the content output in this way is of high quality and very close to your expectations. If you want to fully meet your expectations, you only need to slowly add your requirements.
A small amount of data can hardly serve as a knowledge base because the sample is too small. When your questions are more numerous and complex, it can no longer meet your requirements. How to easily obtain high-quality content will become your second desired tool, and this tool is called a crawler. This is one of the most basic and essential tasks for us AI practitioners.
Here we provide beginners with a newbie starter pack so everyone can easily complete tasks such as downloading, scraping, and uploading. The corresponding tutorials are all written.
[baike]Data Scraping Newbie Starter Pack[/baike]
[baike]Image Generation AI[/baike]
Image generation falls into two major categories. One is having the machine draw and generate images through prompts; the other is extracting and displaying images from an image library. The latter is almost meaningless, and we will only briefly mention it here. Mainstream AI image generation tools include ChatGPT’s DALL·E 3, Midjourney, Leonardo AI, and Stable Diffusion. After 2023, ChatGPT’s DALL·E 3 achieved an overwhelming victory in market competition. The reason is that ChatGPT allows everyone to effortlessly turn words into images through simple natural-language input and output, while the other three require their built-in complex settings and commands to achieve this.
Experience and Installation
ChatGPT’s DALL·E 3 is integrated into the Plus membership. It can be used with the paid version of ChatGPT-4 and custom GPTs. The current cost is 140 yuan/month. Midjourney also costs 140 per month. Leonardo has a certain free quota and is very similar to Stable Diffusion. Stable Diffusion is different from what was introduced earlier: it is a PC-installed version and requires a relatively strong computer configuration. For the graphics card, at least 4 GB of VRAM is needed for smooth use. It also requires a large amount of memory space, roughly dozens of GB. But the advantage is that Stable Diffusion is a free open-source project. At the same time, many technical personnel continuously provide plugins to add personalized functions.
Recommended Related Tools
To make it convenient to display and use, we have organized all tool addresses into a navigation webpage. In the later software section, we also added online tools such as online PS, image upscaling, and image outpainting. We recommend bookmarking it. Usually, the moderators of this site keep this open all the time: https://www.niuboyi.com/123/ We use the navigation site as our workbench. This is also the third category of tools we recommend to you.
For the installed version of Stable Diffusion, please go to the corresponding section to install, download, and study the tutorial: https://www.niuboyi.com/forum-14-1.html
Currently, there are not many good comic tools available across the internet. The chief moderator has developed a ChatGPT comic tool that you can use as a reference: https://www.niuboyi.com/thread-451-1-1.html
Difficulties in Image Generation
Both image and video generation have major difficulties. For people who are not professionals in this field, drawing images is a cross-disciplinary activity. To draw images that meet expectations requires professional knowledge. Whether it is a single image or continuous images such as comics, the elements and descriptions in the entire composition need to be set. Non-professionals do not understand this, so you may be unable to describe it, and ultimately unable to draw images that meet expectations.
For example, prompt: generate a comic character, male, 18 years old, Chinese. This is highly random. You have not specified hairstyle, hair length, face shape, skin color, height, or weight. What is even harder is the overall visual style, such as painting style (color tone, Studio Ghibli imitation, etc.), drawing method (hand-drawn, illustration, sketch), and shot (medium, long, close-up). If you can only give a simple description, it means every time you use this command to generate a character, the result will be different.
Image AI generation software all have a concept of a seed value, which represents the seed number. The meaning of this seed is that it represents the description of the image. The basic technique for generating consistent characters is to make it refer to the original character’s seed number when describing the character’s actions, which makes it easier to generate consistent characters. But remember, the picture composition is not just the character; scene description is also a project. We developed the comic tool with these difficulties in mind—you can imagine how complex it is. https://www.niuboyi.com/thread-451-1-1.html
[baike]Video Generation AI[/baike]
The upcoming Sora will simplify the process of image generation consistency. From the current official Sora operation experience panel, we can compose scenes by simply clicking fixed configurations, such as character + scene + weather + action + time, etc. Sora is currently the most anticipated video software. Other video generation tools such as Pika Labs and Runway generate very short and blurry videos, only a few seconds long. Sora is overwhelming and disruptive. Therefore, we will not elaborate further on other AI video tools.
To learn about Sora, see here https://www.niuboyi.com/forum-53-1.html
Video Frame Elements
The visual elements of video are the same as image composition elements. Friends who understand the concept of frames know that video is also composed of frame-by-frame images linked together, like a film projector showing a flipbook. Video has dynamic motion effects in addition to images, so it has an additional concept of time and space compared with images. In addition to the image elements mentioned above, in the video generation process you need to consider the chained changes of position, time, and weather. Video and comics are of the same nature—both are moving images.
Getting Started Quickly
Start with a variety of image and comic tools, continuously generate images and comics, try different visual effects and styles, generate stories, understand where the problems are, and accumulate experience and understand the difficult issues
Study our tutorials and learn how to create continuous characters with consistent characters and consistent style, and how to control visual style when creating characters, etc.
Course registration address: https://www.niuboyi.com/fidtype-53-type-8-1.html
Recommended Related Tools
CapCut—for video post-production, this one is enough. The latest version 5.5 also adds voice cloning. You can record several different voices from different people, and then dub different characters in the story very well. CapCut also integrates one-click outfit changes and one-click video generation from images based on copy. It is simple and easy to use. It is a very strong video software. Douyin’s entire system is also researching AI, and it has many outstanding performances internationally. At present, it should be considered the strongest in China.
Digital Human Generation
Video generation also includes a category: digital humans. Mainstream software includes DID, heygen, Xiaoice, etc. At present, the moderators tend to recommend heygen, which has a free quota and is also very powerful. Xiaoice is also good, but paid versions of digital humans are all very expensive. For a simple trial, just use the free quotas of these three. They are all in our website navigationhttps://www.niuboyi.com/123/
What Can Digital Humans Do?
Livestreaming and video generation. Combined with voice cloning and ChatGPT’s Q&A function, it becomes a complete system. However, the entire development of this system requires technical integration, and it is too difficult for individuals to operate. An expert on Bilibili has achieved this; you can refer to https://www.niuboyi.com/thread-313-1-1.html
[baike]Voice Generation AI[/baike]
Voice generation includes speech, background sound, songs, and sound effects. All can be generated with AI. Speech is the most important. Once speech is solved first, the rest can be very simple. Previously we used CapCut, iFlytek, Magic Voice Factory, etc. Now these methods are about to become outdated, because cloning software is currently the mainstream. It can imitate emotional tones, and the voice is hard to judge as real or fake. It can even use the voices of celebrities and experts.
Recommended Speech Cloning Tools
Here we recommend two: CapCut and gpt-sovitas. CapCut requires payment and can only record samples on-site; you cannot use other people’s voices. gptsovitas is very flexible and can clone using other people’s voice samples, but it also requires a relatively high computer configuration, also requiring a GPU—that is, graphics card configuration. 4 GB of VRAM is the baseline, but the effect is poor. We suggest that students doing AI should have a computer with a dedicated NVIDIA GPU with 8 GB of VRAM. If you are unwilling to invest, you can buy services outside to have someone train it for you or use a free cloud server.
For CapCut, just use Baidu’s download area. For gptsovitas, you can learn to install it on this site: https://www.niuboyi.com/forum-49-1.html
Recommended Music Generation Tools
We also recommend two: one is svcsovitas, and the other is stableaudio. The former is music cloning; last year’s very popular AI Stefanie Sun was made with this tool. The latter was created in cooperation with Stable Diffusion, and currently many people on NetEase are using it to create original music and make money.
svcsovitas https://www.niuboyi.com/thread-231-1-1.html
stableaudio https://www.niuboyi.com/thread-480-1-1.html
Sound Effect Generation Tools Not Recommended
Here we do not recommend sound effect generation. We suggest you directly download packs online, or spend a few yuan on Taobao to buy a comprehensive sound effect pack; that is enough. There is no need to spend a lot of time and energy generating them yourself.
[baike]Tutorial Summary[/baike]
This issue aims to provide beginners with a comprehensive AI introduction. The content is still fairly basic, but as an introduction and guide to understanding AI, it is enough. If you still do not understand, we suggest reading it twice more. If you still do not understand, come and contact us for private tutoring. |
了解AI 玩转AI
Learn AI in One Article: Essential Skills and Must-Have Software for AI Professionals to Help You Get Started Quickly and Improve Work Efficiency
|NIUBOYI
( 京ICP备17025393号 )|网站地图