[Sora功能介绍] What knowledge do you need to make good use of Sora? Prepare these in advance and you’ll get started with ease.

After Sora's public beta, it doesn't mean you can use it well right away. If you are smart enough, spending 3 months studying it might let you succeed. Drawing on his experience in comic plot creation with Dalle3, the moderator tells you what knowledge you need to do story content generation well.

1. Human-computer interaction experience that people new to AI need to have

Any OpenAI product uses natural language. Their goal is to abandon formatting and commands so that ordinary people can use it well. But it was designed by foreigners, and it understands and outputs in a foreign style. Therefore, you need to learn how to do human-computer interaction and control correct input and output; otherwise, the cost of ineffective communication is too high.

If you can't use it well, the process you get is: prompt\understanding\adapting your prompt before passing it on\execution\personalized feedback
If you use it well, the process you get is: prompt\understanding\directly passing your prompt\execution\feedback as you want

Learn to use prompts: http://www.niubo1ai.com/thread-557-1-1.html


2. Understand image composition

Whether video or image, their composition is the same. Video is also made up of image frames, so we must first correctly generate the first image. Then you must clearly describe all the elements in the image to complete the composition you want.

Composition includes characters, scenes, props, weather, lighting, drawing method, camera angle, etc.

If you say "generate a beautiful woman walking by the river for me," then the above elements are all random, and every time you repeat this sentence you will get a different result.

But if you say, "a 20-year-old Chinese woman, short black hair, white T-shirt, gray jeans, white sneakers, walking by the river, modern urban style, hot weather, 40 degrees, harsh light, irritable atmosphere, medium shot," although this description is still not very complete, the generated video will be much closer to the final result you want.


3. Achieve consistency

A 1-minute video can hardly satisfy the audience's appetite, so making videos is continuous creation, not a single continuous shot. After the audience finishes watching, they need a sequel; you can't change the characters and style in the next episode. Therefore, we need to achieve:

Character consistency How Sora keeps characters unchanged and how to design consistent characters
Style consistency How Sora keeps style unchanged and how to design consistent style
Scene consistency How Sora keeps scenes unchanged and how to design consistent scenes
Visual consistency and camera usage, these two haven't been written yet

Principle of consistency: http://www.niubo1ai.com/thread-544-1-1.html


This knowledge already involves professional knowledge of painting, comics, and film/TV shooting. The moderator's previous major was multimedia design, and I happened to have studied all these things, so I have the opportunity to share this knowledge with everyone.