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(バイク女子編)

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！
ということで、今回は、バイク女子の動画で試してみました。

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、ChatGPTで生成しました。
バイクは「ハーレーダビッドソン XLH883」と指定しました。
道の駅とだけ指示して、場所は特に指定しませんでしたが「道の駅　しらやまさん」になりました。

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。

複雑な動作をする動画のプロンプトは、さすがにチャッピーさんでも一発ではイメージ通りにならず、10回以上やりとりをして作ってもらいました。

A realistic cinematic video based on the reference image.

A young Japanese woman in a black tight leather motorcycle suit stands on the left side of a Harley-Davidson XLH883-style motorcycle at a rural Japanese roadside station parking lot during daytime. The background shows a michi-no-eki building, parked cars, signs, and mountains.

Important continuity rule: after she puts on the black full-face helmet, it must remain firmly on her head until the end of the video. She never removes it. She must not ride without a helmet.

She smiles at the viewer, waves her right hand, and says in Japanese, “またねー”, in a cute, sweet, youthful feminine voice.

She then puts on the black full-face helmet securely with both hands, adjusts it, and fastens it under the chin. The helmet stays on her head clearly.

Now she mounts the motorcycle slowly and clearly. She stands on the left side of the motorcycle. Both hands hold the handlebars. Her left foot is firmly planted on the ground beside the motorcycle, supporting her body and helping balance the bike. Her left leg does not lift.

Only her right leg moves. She shifts her weight onto her left leg, leans slightly forward, and swings her right leg high and wide from the rear wheel side, passing it over the seat in a large smooth arc. Her right boot clears the rear seat and lands on the right side of the motorcycle. She then sits down naturally in the saddle, astride the motorcycle.

After she is fully seated, both feet touch the ground. She remains fully helmeted. She holds the handlebars with both hands in a stable riding posture.

Then she squeezes the left clutch lever with her left hand and presses the starter button with her right thumb. The Harley engine starts with a deep V-twin Potato Sound: “potato-potato-potato,” heavy, bassy, throbbing, and low-frequency.

She lightly twists the right throttle. The engine revs with a deep bass “vroom… VROOM,” then returns to the Potato Sound idle.

She looks ahead toward the parking lot exit. The motorcycle begins to move forward smoothly and slowly. She rides away fully helmeted as the deep Harley Potato Sound fades into the distance.

Photorealistic, natural motion, realistic motorcycle operation, stable wide camera, clear mounting motion, no subtitles, no extra people, no distorted hands, no duplicated motorcycle.

Negative prompt:
No lifting the left leg during mounting, no left-leg swing over the bike, no wrong mounting motion, no missing helmet, no disappearing helmet, no helmet removal, no riding without a helmet, no bare head after wearing the helmet, no prop disappearance, no high-pitched sportbike sound, no electric motor sound, no scooter sound, no subtitles, no floating text, no extra limbs, no broken fingers, no wrong clutch hand, no wrong throttle hand.

共通の設定
width: 1280
height: 704
length: 241 (10秒)
fps: 24

30本以上生成して、その中でマシなのを選びました。
細かい動作の指示は、なかなかイメージ通りにならないですね。

ノイズシード: 886605022695516
乗り方はOKです。

ノイズシード: 813432975650152
変わった乗り方。

ノイズシード: 50724591016763
なんか惜しい！

ノイズシード: 828669917245666
バタバタ…

ノイズシード: 93990002280857

またねー！