LTX-2.3の動画生成プロンプトはチャッピーさん(ChatGPT)に作ってもらおう(男女の会話編)

LTX-2.3の動画生成プロンプトは、チャッピーさん(ChatGPT)におまかせ！
ということで、今回は、登場人物が二人の場合で試してみました。

こちらの記事も参考にしてください。

スタートフレーム用の画像は、以前も登場した、架空の俺と、架空の俺の嫁です。(ChatGPTで生成した画像です。)

ComfyUIのLTX-2.3の動画生成ワークフローは、こちらを使用しました。

会話のタイミングなど、なかなかイメージ通りにならず、チャッピーさんと10回以上やりとりをして、プロンプトを作ってもらいました。
頑張って自分で考えても、こんな複雑なプロンプトは、なかなか作れません。

Positive prompt:
A loose Japanese comedy skit in a normal home living room.

A young Japanese man and a young Japanese woman sit side by side on a sofa.
The man is on the viewer's left.
The woman is on the viewer's right.
Both look relaxed, casual, and slightly lazy.
The overall mood is loose, silly, low-energy, and awkwardly funny.

At first, both are seated on the sofa.

During the "よめ～" and "おれ～" exchange, they do not look at the camera.
They look at each other's faces.

First, the man turns his head toward the woman, looks at her face, gently pushes her shoulder, and lazily says: “よめ～”

Next, the woman turns her head toward the man, looks at his face, gently pushes his shoulder, and lazily says: “おれ～”

Next, the man again looks at the woman's face, gently pushes her shoulder, and lazily says: “よめ～”

Next, the woman looks at the man's face and replies brightly and shortly: “はい！”

Only after the "Hai!" line, both turn their faces toward the viewer.
Then both silently stand up from the sofa with both arms lowered.
No cuts. No interruption.

After standing, both make a fox sign hand gesture, also known as a malocchio sign or metal horns hand sign.
Each hand has the index finger and little finger extended, while the middle finger and ring finger are folded down.
The thumb lightly holds the folded fingers.

Then both clearly cross their arms in front of their chest.
This crossed-arm pose is the starting pose for the dance.

Both begin one continuous chant together in a fast but lazy rhythm:
“ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ ジャンガ、ジャン、ジャジャーン”

The most important motion is the arms opening and closing.
The large arm movement is more important than the exact finger shape.

During the first 6 "ジャンガ" beats, both slowly crouch down.
At the same time, both clearly open their arms outward from the crossed position.
Their arms move from crossed in front of the chest to wide open left and right.
They try to keep the fox sign hand gesture while opening their arms.
By the 6th "Janga", both are in a low crouching pose with both arms fully spread wide.

During the next 5 "ジャンガ" beats, both slowly stand back up.
At the same time, both clearly close their arms inward from the wide-open position.
Their arms move from fully spread wide back to crossed in front of the chest.
They try to keep the fox sign hand gesture while closing their arms.
By the 5th "Janga" after the crouch, both are standing upright with their arms crossed again.

At the final word "ジャンジャジャーン", both hold the final standing crossed-arm pose.

Every "ジャンガ" beat has a clearly audible synchronized guitar strum.
The final "ジャンジャジャーン" has a clearly audible dramatic guitar chord at the exact same time.

Performance style:
Loose, silly, lazy Japanese comedy mood.
The chant is fast, but the body movement is lazy and slightly sloppy.
The arm movement must be large, clear, and easy to see.
The arms must visibly open while crouching and visibly close while standing up.
The fox sign hand gesture should be visible if possible, but the arm opening and closing motion is the top priority.

Camera and framing:
Stable front-facing camera.
Medium-wide shot.
Keep both people visible from head to knees or full body.
Single continuous take.
Clean natural live-action home video.
Raw ordinary camera footage.
The frame stays simple, realistic, and uncluttered.
The screen shows only the two actors, the sofa, and the living room.
Natural realistic movement and ordinary home-video visual style.
No camera cuts.
No scene cuts.

Negative prompt:
romantic mood, flirting, kissing, hugging, sensual intimacy, faces moving very close together, dramatic stage show, music video style, anime style, manga style, poster design, title card, end card, logo, watermark, UI elements

当初、意味不明な字幕が出るのを抑制するのに苦労しましたが、チャッピーさんとやりとりをして、ショート動画は字幕が出やすいということが分かりました。
プロンプトで「ホームビデオ」だと指示することで、字幕が出なくなりました。
プロンプトの以下の部分です。

Clean natural live-action home video.
Raw ordinary camera footage.
The frame stays simple, realistic, and uncluttered.
The screen shows only the two actors, the sofa, and the living room.
Natural realistic movement and ordinary home-video visual style.

共通の設定
width: 1280
height: 960
length: 241 (10秒)
fps: 24

50本以上生成して、その中でマシなのを選びました。
それでは、どうぞ！

ノイズシード: 553953090968218

ノイズシード: 414599301105271

ノイズシード: 512144246003568

意外に難しかったジャンガジャンガ…