AI 動画にセリフとリップシンクを加える方法
映像と同時に生成するセリフ、別に作る音声、ナレーションの3つのルートで、AI 動画にセリフとリップシンクを加える方法を学びます。そのうえで、俳優を演出するように合成音声を演出します。
このレッスンで学ぶこと
- AI 動画でセリフを扱う3つのルート：ネイティブ、別に作る音声、ナレーション
- プロンプトにネイティブのセリフを書く方法：言葉、言語、話し方
- キャラクターの Add Voice ボタンとオーディオノードで、音声ファイルを ScreenWeaver に取り込む方法
- 文章で書いた演技メモで合成音声を演出する方法
- ナレーションが多くの場合、最も簡単で最も映画的なルートである理由
- 実在の人物の声をクローンする際の同意のルール
セリフは AI 映画が最も破綻しやすいところです。声が違う、タイミングがずれる、口の動きが合わない。このレッスンでは、AI 動画でセリフを扱う3つのルートと、それぞれの使いどころを解説します。ネイティブのセリフは映像と一緒に生成され、声、口の動き、環境音が同じテイクに収まります。The Photo の Thomas のセリフはこの方法で作られました。別に作る音声は、先にテキスト読み上げ（TTS）モデルで作るもので、演技とタイミングをコントロールできます。ScreenWeaver ではこれをオーディオノードに入れ、音声を受け付ける動画モデルに送ることができます。ナレーションならリップシンクはまったく不要です。Lost Garden は、顔も口もない騎士たちでこの発想を活かしています。さらに、演技メモで合成音声を演出する方法、各キャラクターの声を同じ人物の声として保つ方法、そして音声クローンにおける同意のルールも学びます。
全文の文字起こし
英語ナレーションの文字起こしです。
0:04Spoken lines are where AI films break most often – the wrong voice, the wrong timing, lips that don't match. In this lecture you'll learn the three routes for dialogue, when to use each one and how to direct a synthetic voice like an actor.
0:19Route 1 – native dialogue, generated with the picture. Some video models produce the voice, the lip movement and the ambience in the same take. That's how Thomas' line in the photo was made. The words are in the prompt, in quotes, with the language and the delivery, softly. Keep native lines short, one or two sentences, and keep the speaker clearly in frame.
1:02Route 2 – a separate voice, then the picture. You generate the voice first, with a text -to -speech model, so you control the performance and the timing, then the voice drives the video. In Screenweaver, a character's story element has an Add Voice button. It creates an audio node, where you upload the voice file, and that audio can be sent as a reference to the video models that accept audio. Dedicated lip -sync tools exist, too, outside the app, for shots that need frame -perfect sync.
1:31Note that the app doesn't generate voices itself – you bring the audio file. For a short film, a text -to -speech service with voice design is usually enough. For recurring characters, save the voice settings of each one, so every scene sounds like the same person.
1:46Whichever read you choose, direct the voice like an actor. Write an acting note in words – calm, grounded, a little tired, a southern English accent. The narration of this course was generated exactly that way, from a single voice and a single acting note. Keep one voice per character for the whole film. Generate a full scene in one pass, so the time beholds, and listen for mispronounced names before you move on.
2:12For Thomas's line, the note could be quiet, almost a whisper, a man who has rehearsed this sentence for days and still isn't ready to say it. A few words of direction change everything about a line.
2:26Route 3 – voiceover, with no lips to sync at all. An inner monologue, an orator, a letter read over images. It's the easiest route, and often the most cinematic one. Lost Garden uses a version of this idea. Its nights have no faces and no mouths, so there is nothing to sync. Emotion comes from posture, light and voice.
2:54One rule with no exception – never clone a real person's voice without their written consent. And check the rules of the platform where your film will play – some require you to disclose synthetic voices.
3:09Your exercise? Pick one line of dialogue from your script and produce it two ways – native, inside the generated shot, and as a separate voice with an acting note. Compare them in the cut. In the next lecture, we'll build the full soundtrack, edit the film and export it.