AI로 로그라인 쓰는 법: 생성하고, 거르고, 다시 쓰기
AI로 로그라인 쓰는 법: 20개를 생성하고, 엄격하게 거르고, 3개를 남기고, 마지막 문장은 손으로 씁니다. 정확한 프롬프트와, AI 영상이 실제로 찍을 수 있는지 보는 필터까지 담았습니다.
이 강의에서 배울 내용
- 로그라인이 이후 모든 결정의 나침반인 이유
- 단편영화에 맞게 줄인 간단한 로그라인 공식
- AI로 로그라인 20개를 생성하는 프롬프트와 제약 조건이 중요한 이유
- AI 로그라인을 3개로 거르는 세 가지 질문
- AI를 비평가로 쓰고, 최종 버전은 직접 쓰는 법
- AI만의 필터: 영상 모델이 실제로 찍을 수 있는가?
로그라인은 영화가 무엇에 관한 것인지 말하는 한 문장이며, AI 영화 제작에서는 이후의 모든 샷을 결정합니다. 이 강의에서는 씨앗, 즉 머릿속에서 떠나지 않는 몇 개의 이미지에서 출발해, AI를 글쓰기 파트너로 삼아 장르, 톤, 시점을 바꿔 가며 로그라인 20개를 생성합니다. 그다음 세 가지 엄격한 질문으로 거르고, 클리셰를 지우고, 3개를 남긴 뒤, AI를 관객이 무엇을 예상할지 묻는 비평가로 바꿉니다.
최종 로그라인은 손으로 쓰고, 친구에게 테스트하고, 필터 하나를 더 거칩니다: 영상 모델이 실제로 찍을 수 있는가? 강좌 프로젝트가 그 이유를 보여 줍니다. The Photo의 초기 버전에서 Thomas는 봉투에서 열쇠를 꺼냈는데, 모델들이 그 작은 동작을 계속 실패해서 아이디어를 다시 썼습니다.
전체 스크립트
영어 내레이션의 스크립트입니다.
0:04Every film starts as a vague feeling. Siblings. An old photograph. A car in the rain. You can feel there's a film in there, but you can't say what it is yet. And that's the real problem. Not writing. Deciding what your film is about. That decision fits in one sentence, and it's called a logline. In this lecture, you'll use AI to explore dozens of options in minutes, and then choose like a writer.
0:34So why one sentence? Because a logline is your compass. It tells you who the film is about, what they want, what stands in the way, and what's at stake. Every later decision checks against it. The look, the cast, every single shot. If you can't say your film in one sentence, your shots won't know what they're for.
0:58Here's a simple formula. When something happens, a specific character must do something, or else they lose something. For a short film, think smaller. One character, one moment, one change. Here's the logline of our course project. Parked in the rain, on a quiet street, a man hands his sister the one photograph that proves they were once close, hoping it will say what he never could.
1:26Now, where does AI come in? Start with a seed, the few images you can't stop thinking about. Then ask the AI for 20 loglines, each from a different angle. A drama, a thriller, a comedy, a ghost story, the same scene told from the sister's point of view. You're not looking for the answer yet. You're mapping the territory, faster than you ever could alone.
1:52Here's the kind of prompt I use. Here is my seed, a brother and a sister, an old photograph, a parked car at night, rain. Give me 20 loglines for a three -minute short film. Vary the genre, the tone, and the point of view. One sentence each, under 30 words. Notice the constraints. A length, a format, a range. Constraints are what turn AI output from noise into options.
2:22Then you filter hard. Three questions. Can it be shot with very few characters and locations? Does someone change by the end? And is it yours? Does it touch something true for you? Be ruthless with the generic ones. The first answers of a language model are the average of everything it has read. That average has a name. Cliche. Keep three loglines. Cross out the rest.
2:52Let's try it on three of the AI's answers. A grieving brother discovers a magical photograph that lets him travel back in time. That's a cliche, and it needs effects and locations you could never keep consistent. Cross it out. A sister learns her brother sold their mother's house. Strong conflict, but it's told, not shown. Keep it as a maybe. A man hands his sister an old photo in a parked car. Small, visual, and something changes. That one goes to the top.
3:26Next, turn the AI into a critic. Paste your three loglines and ask, What's the cliche here? What would the audience expect, and how could we surprise them? Which one has the strongest ending? A good sparring partner doesn't write for you. It asks the questions you are avoiding.
3:46And then you write the final version yourself. By hand. Read it out loud. Look for the one specific detail that makes it yours. Two children on a beach, seen from behind, is worth more than ten adjectives. Never paste the machine's sentence as your final logline. Your voice is the only thing in this process that no model can generate for you.
4:10Then test it on someone. Read your logline once to a friend. Then ask two questions. What is the film about, and how do you think it ends? If they can't picture a single image, add a concrete detail. If they can guess the ending, you've just found your next problem to solve.
4:29One more filter, and this one is specific to AI filmmaking. Can the models actually shoot it? Every character needs a reference sheet. Every location has to stay consistent. Every object a character handles is a risk. In an early version of our project, Thomas pulled a key out of an envelope. The video models kept failing at that small gesture. So we rewrote the idea. The photograph is already in his hand. Write for what the camera can deliver.
5:00Keep a constraints list right under your logline. The number of characters, locations and key props, and anything risky to shoot. For the photo, two characters, one car, one photograph. No crowd, no action, one line of dialogue. That list will follow you through the entire course.
5:22Here's your exercise. Write down your seed. Generate 20 loglines with AI. Filter them down to three. Let the AI attack them, and write your final logline by hand. You'll find the logline worksheet in the resources. In the next lecture, we'll turn that one sentence into a structure. The beats of a short film that works in three minutes.