すごいインプ、、、
みんなもやりたかったんですね!
プロンプト置いておきます。お好きなキャラクターでどうぞ(少し実写になりがちかも)
----------------------
Based on the uploaded reference character, generate a live-action scene inside a wide Tokyo girl’s apartment — a bright, lived-in one-room that feels almost 1LDK in scale. White walls, warm wooden floor, beige curtains, a low bed with soft bedding, a desk with cosmetics, bookshelves, plants, a standing mirror, a rug, scattered personal items, and a compact kitchen area in the back. The room must have strong depth with clear foreground, mid-ground, and deep background layers.
Place around thirty identical versions of the same character (same face, hairstyle, outfit as the reference) throughout the room, each in a different action or interaction. Vary distance, scale, height, and visibility so the density feels natural.
Foreground (very close to camera / partial occlusion):
- characters walking past the lens, slightly out of focus
- a hand or shoulder entering frame
- one leaning close toward the camera
- one half-visible behind a large plant
- one sitting directly in front, tying her hair
- one kneeling by the desk adjusting objects
Mid-ground (main room area):
- one stretching beside the bed
- one sitting on the bed checking her phone
- one lying belly-down across the bed
- one reaching under the bed
- one organizing cosmetics on the desk
- one flipping through a book on the shelf
- one standing in front of the mirror
- one crouching on the rug
- one leaning against the wall
- one looking out the window
- one adjusting the curtain
- one carrying laundry
- one drinking from a cup
- one tidying pillows
- one sitting on the floor eating snacks
- one doing a small jump or motion blur gesture
- one moving a small chair
Background (deep perspective / near kitchen and hallway):
- one standing near the stove drinking water
- one opening a cabinet
- one sitting on a stool
- one leaning in a doorway
- one walking toward the hall
- one silhouette partly hidden behind the fridge
- one reaching up to a high shelf
- one standing far by the entrance area
- one barely visible through the hallway frame
- one sitting on the floor near the kitchen rug
Ensure strong layered occlusion: foreground characters partially block mid-ground ones, and background characters appear smaller with natural perspective falloff. Scatter the thirty characters organically, avoiding symmetry or grid alignment. Lighting is soft natural daylight from the window, consistent across all characters for full integration.
Placed within a live-action background that matches the illustration’s posture and composition — while faithfully preserving the illustrated texture and style. Realistic lighting, depth of field, and subtle filming effects are applied to blend the illustration seamlessly with the real environment.
AI動画:低空飛行からの直角立ち上がり
地上5cmで180km/h → ギリギリ低空 → 直角上昇 →
ビルの壁面スレスレ加速 → 空へ水平滑空にチャレンジしてみました。
“物理的に無理な軌道”を破綻なく成立させるには、動きを5つのシーンに分解して書くのが一番効きました。
今回のプロンプトで工夫した点👇
① 高速パートと上昇パートを切り分ける
→ 速度変化を同一文に入れると破綻する
② 直角上昇は “snap upward / 90-degree climb” のように短く・強い動詞で明示
→ AIが変化点を理解しやすい
③ 近接描写は “centimeters from windows” で限定的に指定
→ “スレスレ”を理解させつつ環境破綻を防ぐ
④ 加速シーンは “accelerates upward along the building”
→ 方向性を固定してブレを防止
⑤ 最後の水平飛行は “glide / calm / stabilizes” と落ち着いた語彙で減速を伝える
→ 上昇→静寂のコントラストが自然に成立する
10秒の中で“激→静”を成立させるための最短プロンプトがこれ👇
Prompt:
"cinematic ultra-realistic POV sequence, total length about 10 seconds.
Scene 1:
Racing at 180 km/h, only 5cm above the asphalt in a neon city. Strong forward motion blur on the ground, buildings sharp and stable.
Scene 2:
The camera gets even lower, almost scraping the ground, then snaps upward.
Scene 3:
A sudden 90-degree vertical climb between skyscrapers, passing just centimeters from windows and steel edges, no warping.
Scene 4:
Accelerates upward along the building surface, neon reflections streaking across glass, strong parallax.
Scene 5:
Clears the rooftop and shifts smoothly into a calm horizontal glide in the open sky. Speed decreases, view stabilizes with soft haze and cinematic lighting."
■手書き風文字画像の作り方
ChatGPTを活用して、エモい手書き風文字の画像を生成する方法をご紹介します!
▼作り方
①背景になる画像を1枚用意(無くてもOK)
②ChatGPTで①の画像を参照し、下記プロンプトを入力
===
On this image, overlay only Japanese handwritten text. Use exactly the Japanese phrase that I write between the corner brackets 「」 at the end of this prompt, and reproduce it exactly as written, including punctuation. Do not translate, reorder, or omit any characters. Do not split the phrase into multiple separate parts or different lines by yourself; treat it as one continuous phrase.
Place the phrase as handwritten text on top of this image, in an area with enough empty space so it is easy to read. Write it with a loose diagonal flow rather than a perfectly straight horizontal baseline.
Use a thin, single-line pen-like appearance, like a fine gel pen or sign pen, but make the strokes lively and imperfect: slight wobble, small variations in pressure, and natural shakiness so it clearly looks like human handwriting, not a clean font. The style should feel simple, mature and a bit emotional, not cute. Avoid rounded “kawaii” shapes. Make the characters slightly tall and narrow, and allow noticeable variation in size and tilt between characters, as if written quickly in one spontaneous motion. Letter spacing should be irregular, with some tight clusters and some wider gaps, and the baseline should be uneven and relaxed, not perfectly aligned or centered.
The ink color is pure white. Do not add outlines, shadows, gradients, or 3D effects. Do not modify or blur the original image itself; only overlay the handwritten text.
Use this exact phrase:
「ここに好きなフレーズ」
===
背景画像がない場合は、↑のプロンプトの末尾に「背景は透過(or白背景に)してください。」などを入力してください◎
これで手書き風文字画像の完成です!
ちなみにNanoBananaやSeedream4.0だとできませんでした🥲