AI アニメシリーズのケーススタディ：Lost Garden の作り方
AI アニメシリーズ Lost Garden が20分のエピソードを通して一貫性を保つ方法を、制作資料から直接見ていきます。顔のない騎士、世界観バイブル、プロンプトのルール。
このレッスンで学ぶこと
- Lost Garden の第2話が、1つのアイデアと3つの時計の上にどう組み立てられているか
- 顔のない騎士が、AI 映画制作で最も難しい一貫性の問題をなくす理由
- 場所、物、人物のバイブルとエスタブリッシングプレートが、どのように世界を定義するか
- 第2話のあるショットの実際のプロンプト（声の演出つき）
- Lost Garden のすべてのプロンプトで繰り返されるルール一式（声のルールを含む）
- DaVinci Resolve での編集とグレーディングが、各エピソードをどう統一するか
Lost Garden は、Frank Houbre がたった1人で、すべて ScreenWeaver の中で AI を使って制作したダークファンタジーのアニメシリーズで、各エピソードは約20分です。このケーススタディでは制作資料を開き、この講座のメソッドがシリーズの規模でどう機能するかを見ていきます。第2話は1つのアイデア、つまり本当の怪物は絶望だということと、物語を引き締める3つの時計（タイムリミット）から始まります。2人の騎士は中が空洞で顔のない鎧なので、あらゆる感情は身体の動き、首のかしげ方、姿勢、そして光から生まれます。世界は場所、物、人物のバイブルの中に息づき、ロケーションはエスタブリッシングプレートとして生成されます。すべてのプロンプトで同じルールを繰り返し、すべてのエピソードは DaVinci Resolve で編集され、同じ冷たい青にグレーディングされています。
全文の文字起こし
英語ナレーションの文字起こしです。
0:04Everything in this course comes from real productions. Here's the biggest one. Lost Garden, an animated dark fantasy series made with AI, with episodes of around 20 minutes. Let's open the production files and see how the method holds at that scale.
0:21It starts with an idea, not with images. The working summary of the second episode opens with one line. The real monster of this episode isn't a creature, it's despair. Then three clocks tighten the whole episode. Lantern's light leaking out through a hole in his armour, the black masses advancing toward the village by dawn, and the blacksmith who can save him slipping in and out of madness. Structure first, always.
0:49Then the design decision that makes the series possible. The two knights, Lantern and Sarua, are hollow, faceless armours. No face, no mouth, no lip movement. Every emotion comes from body language, head tilts, posture, and light. That single choice removes the hardest consistency problem in AI filmmaking, and gives the series its identity.
1:12The world is written down before it's generated. A bible of places, a bible of objects, a bible of the people who live underground. And the locations are generated first, as establishing plates before the scenes that happen in them. The leaning castle, pierced by a root. The gorge with no bottom. The village hanging over the void.
1:33Then one prompt per shot, and a set of rules repeated in every single prompt. No music. Cinematic 2D anime animation in a precise style. A dynamic camera angle, with poetic framing. The knights are faceless armours. Lantern never speaks, only faint metallic sounds. Sarua always speaks in a warm, playful voice. Those rules are the visual bible, made executable.
2:01Here's a real line from shot 5 of the second episode. Saruso. Warm and tender, but still playful. Your light is your life, little lamp. The tone of the voice is written in the prompt, every time he speaks.
2:18And finally, the edit. Every episode cut and graded in DaVinci Resolve. The shots pulled into the same cold blue, music and sound built in post. Look at the result. The same two nights, minute after minute. Not because a model is magic, but because every decision was made once, written down and repeated with discipline. That's the method of this course, at the scale of a series.
2:47What you can take from it, decide your constraints early, write them down and repeat them in every prompt. Faceless heroes, a fixed style, a voice rule. Constraints aren't a limit on your film. They're what hold it together.