如何避免 AI 粗制滥造:劣质 AI 视频的 5 个特征及解决方法
如何避免 AI 粗制滥造:认识让 AI 视频显得千篇一律的五个典型特征及其解药,从变脸的角色到飘忽的镜头。然后用 10 秒钟检查每一个镜头。
在这一课中你将学到
- AI 粗制滥造内容为什么会出现:没人去改的模型默认值
- AI 粗制滥造内容的五个典型特征,以及每一个的解药
- 如何掌控角色、摄影机、故事、连戏和节奏
- 为什么电影节评委和客户首先淘汰粗制滥造的作品
- 在接受每个镜头之前都要做的快速检查
- 课程项目 The Photo 在五个特征上的得分如何
AI 粗制滥造内容很容易感觉到,却很难说清楚。这一课把它说清楚:让 AI 视频显得千篇一律的五个典型特征,以及专业人士针对每一个特征的解药。变脸的角色,用参考设定图解决。飘忽的镜头,靠每个镜头只决定一种运镜(或者不动)来解决。画面很美却言之无物,用剧本和镜头表解决。世界前后对不上,用连戏、视觉设定集和 180 度轴线规则解决。没有节奏,靠按情绪剪辑、分层搭建声音来解决。
你会明白,为什么这些问题没有一个是模型不好造成的,为什么电影节评委和客户首先淘汰的就是粗制滥造的作品,以及如何对每一个 AI 镜头做快速质量检查,这一切都会在 The Photo 上现场演示。
完整文字稿
英语旁白的文字稿。
0:04You can spot AI slop in about two seconds. Even people who know nothing about AI can feel it. Something is off, and they scroll away. In this lecture, we're going to name exactly what that something is. There are five signatures. Once you see them, you can't unsee them, and you'll know how to keep every one of them out of your films.
0:30First, why does slop happen at all? Because a video model is built to turn whatever you give it into a plausible clip. Give it one line, and it fills in everything else. The face, the lens, the light, the timing. Those defaults are the same for everyone. And that's why slop all looks alike. Every signature you're about to see is a default that nobody overrode.
0:55Signature one. The shape -shifting character. In one shot, she's a cartoon. In the next, she's older, her coat has changed color, the wolf beside her is a different animal. Same prompt. Every time. The audience stops believing in her, and when they start believing in the character, they stop caring about the story. The antidote is the reference sheet. One sheet per character, with the face from the front, three -quarter, and profile, and the costume locked, given to the model on every single shot. Section three is entirely about this.
1:30Signature two, the floating camera. Slow drifts, endless orbits, a push -in on everything, because the model moves the camera whenever nobody tells it what to do. Real cinematographers move the camera for a reason, to reveal, to follow, or to get closer to an emotion. The antidote, decide the camera before you generate. One shot, one movement, written in the prompt. Or no movement at all. A static shot is often the strongest choice you can make.
2:02Signature three, beautiful images that say nothing. A lone warrior on a cliff, golden hour, epic clouds, and then what? Slop is a collection of shots, cinema is a sequence. Each shot answers a question raised by the one before, and raises a new one. The antidote is the script, and the shot list. Before a shot exists, you know why it's there, and what it should make the audience feel.
2:31Signature four, a world that doesn't hold together. It's night in the wide shot, and dusk in the close -up. It's raining outside the window, then it isn't. A character sits on the left, then on the right, for no reason at all. The antidote is continuity. The same discipline a script supervisor brings to a film set. One visual Bible, the 180 degree rule, and the habit of checking every shot against the one before it. In the photo, we wrote down where each character sits in every single shot. Thomas at the wheel, Emily in the passenger seat, the back seat empty. That one line in the prompt prevents a whole family of errors.
3:18Signature five, no rhythm. Every clip lasts exactly five or ten seconds, because that's what the model generated. Music from wall to wall, no silence, no sound design. The antidote, cut on emotion, not on clip length. Keep two seconds of a shot if two seconds is what it needs. Give the audience room to breathe, and build the sound like a real film. Ambience first, then effects, then music.
3:48Now look at the pattern. None of these problems comes from a bad model. They come from decisions nobody made. AI slop is what you get when the machine makes the decisions for you. AI cinema is what you get when you make them yourself and use the machine to execute. That's the shift this whole course is about, from prompting to directing.
4:12And this isn't only about taste. Festival juries watch a lot of AI films, and slop is the first thing they filter out. Clients are the same. No brand can put a character on screen who changes face between two shots. These five signatures are exactly what separates the films that get selected and the jobs that get paid from everything else.
4:34So how do you use this day to day? Every time a shot comes back from the model, run it against the five signatures before you accept it. Is it the same character? Is the camera doing what you asked? Does it match the shot before? If not, reject it fast, and regenerate with the decision you forgot to make. It takes 10 seconds per shot, and it's the fastest quality upgrade you'll ever get.
5:02Let's try it right now, on the photo. Signature one. Is it the same two actors in every shot? Yes. Signature two. One camera decision per shot. A dolly in, two static shots, and a dolly in to finish. Signature three. Does every shot move the story forward? The silence, the photo, the memory, the tears. Signature four. Thomas stays at the wheel, Emily stays in the passenger seat, and the rain never stops. Signature five. The cuts land where the emotion turns. Five out of five. Watch it once more with that in mind.
6:04Here's a quick checklist to run on every film you make. Are my characters the same in every shot? Does every camera move have a reason? Does every shot have a purpose in the story? Do the light, the weather, and the positions match from shot to shot? And does the edit breathe? You'll find this checklist as a PDF in the resources, and we'll use it again on your final project. Your exercise for this lecture. Pick any AI video in your feed today, and score it against these five questions. Then do the same with a film you love. You'll start seeing the difference everywhere.
6:41That's the end of section one. Take the short quiz to lock it in, and then we'll start where every film starts with the story. Next up, from idea to logline with AI as your writing partner.