C.W.K.
Stream
Lesson 09 of 10 · published

Iterative Editing Beats the One Perfect Prompt

~16 min · control, editing, l9

Level 0Spark
0 XP0/100 lessons0/14 achievements
0/200 XP to next level200 XP to go0% complete

피파 한 줄 정리: '완벽한 한 prompt' 대신 iteration loop. Generate-first (탐색) vs edit-first (실행) — 실제 프로젝트는 둘 다 써.

Mental model: A photographer doesn't take one photo and submit it. She shoots 200 frames, selects the best 10, crops and adjusts them, retouches a couple, and delivers 3 finals. A sculptor doesn't carve a masterpiece in one cut — they rough out the form, refine, detail, polish. Creative work is inherently iterative. AI image generation is no different, but most beginners don't realize this.

The One-Prompt Myth

There's a widespread misconception that skilled AI artists write one brilliant prompt and get perfect results. The reality? They generate dozens or hundreds of variations, curate ruthlessly, then edit the best ones. The "skill" isn't in writing one magic prompt — it's in the iteration loop:

The Professional Iteration Loop:

  ┌──────────────────────────────────────────────────────────┐
  │                                                          │
  │   Generate ──→ Curate ──→ Refine ──→ Edit ──→ Finalize │
  │      │            │          │         │                 │
  │      │            │          │         │                 │
  │      ▼            ▼          ▼         ▼                 │
  │   Broad batch  Pick best  Adjust    Inpaint,            │
  │   (many seeds) candidates  prompt   outpaint,           │
  │                            & params  composite           │
  │                                                          │
  │   ←─── Loop back as needed ──────────────────────────── │
  └──────────────────────────────────────────────────────────┘

Generation-First vs. Edit-First Workflows

Two legitimate approaches, each with strengths:

Generation-First: Start with broad text-to-image exploration. Generate many variations. Pick the best. Refine with image-to-image. Fix details with inpainting. This works well when you don't have a specific vision — you're discovering what works.

Edit-First: Start with an existing image (photo, sketch, previous generation). Use image-to-image to transform it. Use inpainting for targeted changes. This works well when you have a clear vision — you're executing a specific idea.

Why Iteration Wins

Several factors make iteration superior to single-shot generation:

  • Reduce risk: One generation might fail. Ten generations almost certainly include something usable.
  • Discover unexpected directions: The model sometimes produces surprising results that are better than what you imagined. You can't discover these without generating broadly.
  • Incremental control: Each editing pass adds precision. You can fix one thing at a time instead of trying to control everything simultaneously.
  • Manage complexity: A prompt with 15 requirements will likely fail several of them. But generating with 3 core requirements, then adding the rest through editing, manages complexity effectively.
❌ One-Shot Perfectionism

"A photorealistic portrait of a woman in her 30s with specific facial features, wearing a blue silk dress, holding a glass of red wine, in a dimly lit Italian restaurant with brick walls and candles, shot on a 85mm lens, f/1.4, warm color grade, cinematic composition, no text, no artifacts, perfect hands" → Trying to nail everything in one prompt

✅ Iterative Approach

Step 1: "Woman in Italian restaurant, warm candlelight, cinematic" (broad) → Step 2: Pick best composition, image-to-image with more detail → Step 3: Inpaint to fix hands → Step 4: Adjust color grade in editor → Done

Key Takeaways
  • Professional AI art is iterative: generate broadly, curate, refine, edit, finalize.
  • The "one perfect prompt" is a myth — skill is in the loop, not the single shot.
  • Generation-first explores; edit-first executes. Use both depending on where you are in the project.
  • Complex requirements are better handled incrementally through editing passes than simultaneously through one prompt.

External links

Exercise

거의-but-not-quite 사랑하는 image 한 장. 처음부터 regenerate 대신 다른 problem 영역에 inpainting 3 pass. Fresh generation과 비교. 어느 게 더 빨랐나?

Progress

Progress is local-only — sign in to sync across devices.
이 페이지에서 버그를 발견하셨거나 피드백이 있으세요?문제 신고

댓글 0

🔔 답글 알림 (로그인 필요)
로그인댓글을 남기려면 로그인해 주세요.

아직 댓글이 없어요. 첫 댓글을 남겨보세요.