AI Prompt Engineering Complete Guide — Information Retrieval, Image/Video Generation, and Debugging (KR/EN/JA)

Basic: "PostgreSQL의 인덱스 종류를 비교해줘"

Improved: "너는 10년차 PostgreSQL DBA야. B-tree, Hash, GIN, GiST 인덱스의
사용 사례, 성능 차이, 실전 선택 기준을 표로 정리해줘.
프로덕션 환경(1000만 행 이상) 기준으로."

Basic: "Tell me about Kubernetes networking"

Improved: "You are a senior Kubernetes platform engineer with CKA/CKS certifications.
Explain the differences between ClusterIP, NodePort, LoadBalancer, and Ingress
with real-world use cases. Include a comparison table and one kubectl example for each.
Focus on production clusters with 50+ nodes."

Basic: "Dockerについて教えて"

Improved: "あなたは経験豊富なDevOpsエンジニアです。
DockerとPodmanの違いを、セキュリティ、パフォーマンス、
エコシステムの3つの観点から比較表でまとめてください。
本番環境での選択基準も含めてください。"

[Subject] + [Style] + [Composition/Camera] + [Lighting] + [Mood/Color]

Bad example: "고양이 그려줘"

Good example: "창가에 앉아 있는 흰색 페르시안 고양이, 따뜻한 오후 햇살이 비치는
카페 안, 얕은 피사계 심도(shallow depth of field), 따뜻한 톤,
35mm 필름 느낌의 사진, 부드러운 자연광"

Bad: "Draw a robot"

Good: "A humanoid robot sitting in a library reading a book,
surrounded by floating holographic screens, warm ambient lighting,
cinematic composition, shot with 85mm lens, shallow depth of field,
cyberpunk aesthetic with warm orange and cool blue color palette, 8K detail"

Bad example: "猫を描いて"

Good example: "桜の木の下で昼寝している三毛猫、水彩画風、
柔らかい春の日差し、パステルカラー、温かい雰囲気"

[Camera Movement] + [Subject's Action] + [Environment/Background] + [Time/Lighting Changes] + [Mood]

"카메라가 천천히 줌인하면서, 네온 불빛이 가득한 도쿄 골목길을
걸어가는 사이보그 여성. 비가 내리고 있고, 물웅덩이에
네온 불빛이 반사된다. 시네마틱, 슬로우 모션, 블레이드 러너 스타일."

"Slow dolly shot through a misty bamboo forest at dawn,
camera gradually rising to reveal a hidden temple,
volumetric light rays through the canopy, gentle fog rolling,
ethereal and peaceful, Studio Ghibli aesthetic, 4K cinematic."

"カメラがゆっくりとドリーインしながら、桜吹雪の中を
歩く着物姿の女性。夕暮れの柔らかい光、
スローモーション、映画的な雰囲気。"

[Error Message/Symptoms] + [Related Code] + [Environment Info] + [What You've Tried] + [Expected Result]

Bad example: "이 코드 안 돼. 고쳐줘."

Good example: "Python 3.12, FastAPI 0.115에서 아래 코드 실행 시
'RuntimeError: Event loop is closed' 에러가 발생합니다.

[code block]

환경: Ubuntu 24.04, uvicorn 0.30
시도한 것: asyncio.new_event_loop() 추가 → 동일 에러
원하는 결과: 비동기 DB 연결이 정상 종료되어야 함"

Bad: "My code doesn't work, fix it"

Good: "I'm getting a 'CUDA out of memory' error when training a LoRA adapter
on RTX 3090 (24GB). Model: Llama-3.1-8B, batch_size=4, max_seq_len=2048.

[code block]

Env: PyTorch 2.4, CUDA 12.4, transformers 4.44
Tried: gradient_checkpointing=True → still OOM
Expected: Training should fit in 24GB VRAM with LoRA rank=16"

Bad example: "動かない。直して"

Good example: "Go 1.22で以下のコードを実行すると、
'goroutine leak detected' という警告が出ます。

[code block]

環境: macOS, Go 1.22, goleak v1.3
試したこと: defer cancel() 追加 → 変わらず
期待する結果: goroutineが正常に終了すること"

You are a [role].
Explain [topic] from the perspective of [criteria/viewpoint].
Organize in [output format] and follow [constraints].
Include [N] practical examples.

[Detailed subject description], [action/pose],
[detailed environment/background], [camera angle] shot,
[lighting type], [color palette/mood],
[style], [quality keywords]

Environment: [OS, language version, framework version]
Error: [exact error message]
Code: [relevant code block]
Tried: [solutions already attempted]
Expected: [expected behavior when working correctly]