🧱壁に落書きした話
こんばんは。私です😋
先日、KITAcoreさんの記事で素敵なプロンプトに出会いました。
「実写風の人物」と「壁に描かれたちびキャラ」を一枚の画像に共存させるという内容なのですが、めちゃくちゃいい感じでした。（語彙）
これは試してみたい！😤
ということで、早速プロンプトをChatGPTにペタリしました。
👇こちらが実際に使ったプロンプトです。
Use the uploaded character reference image as the strict identity, outfit, and styling reference.
Preserve the reference character's:
- face identity
- facial proportions
- eye shape
- nose
- lips
- skin tone
- hairstyle
- hair color
- visible headwear, eyewear, and hair accessories
- outfit, layering, and footwear shown in the reference image
- overall recognizable vibe
Do not hardcode any traits that are not present in the uploaded reference.
All identity, hairstyle, accessory, and clothing details must be inferred directly from the uploaded reference image.
=== CORE CONCEPT ===
Create ONE single image showing the SAME person twice:
1. a photorealistic version of the uploaded person, standing in front of a wall
2. a chibi deformed illustration, oversized head, simplified rounded features, soft flat colors version of the SAME person, drawn directly ON the wall surface beside them
The drawn figure is artwork on the wall. It is NOT a second real person and NOT a cast shadow.
REAL PERSON (LEFT of the frame):
- fully photorealistic photography, natural skin texture and natural fabric texture
- wears the same outfit, footwear, and visible details as the uploaded reference
- actively participates in the pose in a natural, realistic way, never stiff
WALL DRAWING (RIGHT of the frame):
- rendered entirely in chibi deformed illustration, oversized head, simplified rounded features, soft flat colors
- reads as artwork applied to the wall surface, following the wall's plane and its lighting
- resembles the person through hairstyle silhouette, accessories, outfit silhouette, and pose structure
- slightly taller than the real person
- The drawn figure performs a far more exaggerated, chaotic, cartoonish version of the same idea, while the real person performs a natural, restrained version. The drawn figure should look like it is having far too much fun.
- surrounded by hand-drawn motion lines, hearts, stars, sparkles, and comic effect marks in the same medium
=== POSE ===
Both figures share the same general pose idea but are not identical.
The drawn figure is free to push that idea further.
The pose must NOT be fixed. Invent a fresh pose for each generation.
Do not use pointing poses, finger-gun poses, or a plain standing pose.
=== SCENE ===
pale plaster wall with a soft patch of natural sunlight falling across it.
Keep the wall and the floor uncluttered so both figures read clearly.
=== COMPOSITION ===
- aspect ratio 4:5
- full-body framing, both figures visible head to feet
- the real person on the LEFT side, the wall drawing on the RIGHT side
- keep enough space around both figures so the full pose stays visible
- the relationship between the real person and the wall drawing must be clear at a glance
=== LIGHTING ===
- soft natural daylight, clean and polished
- one consistent light direction for the person, the wall, and the floor
- the person's face must stay clearly visible
=== STYLE QUALITY ===
- photorealistic human photography for the person
- confident, medium-authentic rendering for the wall drawing
- strong identity resemblance between the person and the drawing
- outfit and styling faithfully based on the uploaded reference
- clean composition, social-media-friendly
- no obvious AI artifacts
=== NEGATIVE ===
repeated pointing pose, finger-gun pose, same pose every time, fixed pose, stiff standing pose, boring mirrored pose, realistic second person, a second photographic person, normal cast shadow, physical shadow only, solid black silhouette shadow, horror shadow, creepy shadow, the drawing detached from the wall, floating object in front of the wall, three or more figures, weak resemblance, mismatched outfit between the person and the drawing, outfit change unrelated to reference, hardcoded traits not present in the reference, cluttered wall, cluttered background, mismatched lighting, text, watermark, logo, signature, distorted body, extra limbs, extra fingers, bad hands
よく見ると、見出しごとにきちんと整理されているので、素人ながら「土台部分」と「変えたら楽しそうなアレンジ部分」を分けて考えることができました。
キャラクターを忠実に再現する部分や実写と壁イラストという構成は、このプロンプトの骨格なのでそのまま残しました。
以下の英文は元のプロンプトではなく、私がアレンジで加えた一例です。
SCENE（背景）
=== SCENE === で検索
a cozy cafe with wooden furniture木の家具があるカフェに。
ただし、ちびキャラが描かれる壁だけは残します。
LIGHTING（光）
=== LIGHTING === で検索
cinematic backlighting映画のような逆光に。顔がはっきり見える条件は残しています。
COMPOSITION（構図）
=== COMPOSITION === で検索
low-angle shotローアングルに。
「実写は左、ちびキャラは右」の配置はそのままです。
POSE（ポーズ）
=== POSE === で検索
looking outside the frame画面の外を見る姿に。
実写とちびキャラで動きを揃えると、プロンプトらしさが残ります。
WALL DRAWING（壁のちびキャラ）
WALL DRAWING (RIGHT of the frame) で検索
cheerful and wildly excited思いきり楽しそうな様子を追加。
実写は、自然にちびキャラは大胆に。
この差が大きいほど楽しい画像になります。
ということで、そのあたりを少しだけアレンジしてるとこんな感じになりました。
場所や光、構図を少し変えるだけで、思っていた以上に印象が変わります。
私はこの辺りを中心に変えてみましたが、画像生成は複数の指示が影響し合うので、同じように変えても必ず同じ結果になるとは限りません。
完成度の高い土台があったからこそ、「少しだけ自分なりに遊んでみよう」と思えました。
わりとざっくりしたアレンジでもいろいろ遊べそうなプロンプトなので、みなさんもよかったらやってみてはいかがでしょうか。
あ、実際に壁に落書きをしちゃダメですよ。