SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsmedia-gpt-image2-builder

media-gpt-image2-builder

Design & creation

View repository

Updated 10-15 21:39

About this skill

media-gpt-image2-builder — Day 3 S2 5 Korean typographic ad images

Overview

One-line responsibility: A skill that takes a student's one-line natural language input during Day 3 S2 (11:00–11:50), automatically rewrites it internally into an 8-step prompt, and generates a set of 5 Korean typographic ad images with GPT Image 2.

V6 18 Deliverable Chain Position: Deliverable ⑬ (Day3 S2, 11:08–11:18 instructor demo → 11:18–11:45 student practice) — ⑧ 7-step copy + ⑫ moodboard → ⑬ 5 images → flows into ⑭ composite PNG input.

Student non-disclosure rule (REQ-MEDIA-013 HARD): Students input only one line of natural language. The plugin internally automatically rewrites it into an 8-step prompt and passes it to GPT Image 2. The original 8-step structure is not exposed in the student response.

OpenAI model restriction (REQ-MEDIA-014 HARD): Use only GPT Image 2 (image-only). Calling GPT-5.5 (LLM) or Sora (video) is prohibited.

Trigger keywords

  • "Make 5 ad images", "Generate image set"
  • "Korean typo image", "Hero image", "Info image"
  • "Life image", "CTA banner"
  • "Make it with GPT Image 2"
  • "Day3 S2 image", "/media-gpt-image2-builder"
  • "Product ad image set"

Workflow

Input slots

Slot Source Description
user_prompt Student one-line natural language e.g., "Make 5 mask pack ad images"
copy_7step Deliverable ⑧ 2 sets of 7-step copy (problem awareness, empathy, core, proof, usage, FAQ, CTA)
moodboard Deliverable ⑫ Color palette, tone keywords, reference direction

Internal operation of this skill

Only the one-line natural language input is exposed to students. The 8-step processing below is the skill's internal operation, and the original 8-step structure is not displayed on the student response screen.

Step 1 — Automatic rewriting of natural language → 8-step structure (internal)

Step Content Moodboard·Copy linkage
1) Purpose/use Channel and goal where the image will be used Refer to channel message ⑩
2) Core brief Product USP + target emotion Copy ⑧ Step 3 core message
3) Required elements Logo, text, product placement Moodboard ⑫ tone keywords
4) Context/environment Usage situation, background Persona ⑦ Lifestyle
5) Composition/Space Aspect ratio·Margins·Focus By image type (Hero/Info/Life/CTA)
6) Light/Color/Material Lighting direction·Color palette·Material feel Moodboard ⑫ Color palette
7) Constraints/Prohibitions/Fixed Copyright·Brand guidelines·Prohibited elements Regulations by category
8) Output/Format Resolution·Aspect ratio·File format Channel specifications

Step 2 — Mapping 5 image types

Image Role Copy ⑧ Mapping Aspect ratio
Hero Product-focused·USP delivery Stage 1 (Problem awareness) + Stage 3 (Core) 1:1 or 3:4
Info Ingredients·Proof·Data text overlay Stage 4 (Proof) 1:1
Life 1 Pre-use lifestyle·Emotion Stage 2 (Empathy) 3:4
Life 2 Post-use results·Change Stage 5 (Usage·Benefits) 3:4
CTA banner Discount·Purchase action inducement Stage 7 (CTA) 9:16 or 1:1

Step 3 — GPT Image 2 call (OpenAI API)

OpenAI API: POST /v1/images/generations
model: "gpt-image-2"
size: 이미지 유형별 (1024×1024 / 1024×1792 / 1792×1024)
quality: "hd"
n: 1 (이미지당 1회, 총 5회 호출)

Maintain same seed/style consistency: fixed seed + fixed prompt style prefix

Step 4 — Automatic Korean readability correction (REQ-MEDIA-010)

When GPT Image 2 Korean typography breakage is detected:

  1. Automatic re-instruction: Attempt automatic correction once, e.g. "Thicken Korean font", "Reduce letter spacing"
  2. On failure: Guide text to be handled in a separate compositing step (⑭)

Step 5 — Automatic AI disclosure chain (REQ-MEDIA-002)

After 5 image generation is complete media-ai-disclosure automatically chain to attach metadata + AI generation disclosure.

Pass criteria (PDF §6.8 ⑬)

  • Same seed/look consistency (Style unification across 5 images)
  • Korean readability (No typography breakage)
  • Automatic AI disclosure metadata attachment

Usage examples

Example 1: Beauty mask pack advertisement image

수강생 입력: "마스크팩 광고 이미지 5장 만들어줘"
→ (내부) 8단계 프롬프트 자동 리라이팅 + 무드보드⑫·카피⑧ 연동
→ 결과: Hero("피부가 바뀌는 순간") + 인포(성분 표 오버레이) + 라이프1(세안 후 루틴) + 라이프2(글로우 피부) + CTA(오늘만 30%↓)
→ media-ai-disclosure 자동 체인 → AI 생성 메타데이터 부착

Example 2: Health food collagen advertisement image

수강생 입력: "콜라겐 음료 광고 이미지 세트 만들어줘"
→ Hero(숫자 강조 "98%") + 인포(원료 클로즈업) + 라이프1(일상 한 모금) + 라이프2(피부 결 변화) + CTA(첫 구매 특가)
→ AI 표기 자동 부착

Example 3: Automatic correction of Korean typography breakage

수강생 입력: "생활용품 광고 이미지"
→ (내부) 생성 시도 → Hero 한글 깨짐 감지
→ 자동 재지시: "Bold 체 한글 타이포, 자간 -10%, 두꺼운 고딕" 보정 1회
→ 보정 성공: 5장 세트 반환 / 보정 실패: "⑭ 합성 단계에서 텍스트 추가 권장" 안내

Output format

{
  "image_set": {
    "product": "마스크팩",
    "seed": 42,
    "style": "클린 스튜디오, 화이트 배경, 뷰티 에디토리얼",
    "images": [
      {"type": "Hero", "aspect": "1:1", "url": "...", "copy": "피부가 바뀌는 7일", "ai_tagged": true},
      {"type": "인포", "aspect": "1:1", "url": "...", "copy": "히알루론산 50mg", "ai_tagged": true},
      {"type": "라이프1", "aspect": "3:4", "url": "...", "copy": "매일 밤 10분", "ai_tagged": true},
      {"type": "라이프2", "aspect": "3:4", "url": "...", "copy": "7일 후 달라진 피부결", "ai_tagged": true},
      {"type": "CTA배너", "aspect": "9:16", "url": "...", "copy": "오늘만 30% 할인", "ai_tagged": true}
    ],
    "disclosure": "AI 생성 메타데이터 부착 완료 (media-ai-disclosure)"
  }
}

Pass criteria

Item Criteria
Number of images 5-image set (Hero 1 + Info 1 + Life 2 + CTA banner 1)
Style consistency Maintain same seed/look (Color palette·style unification across 5 images)
Korean readability No typography breakage (Including 1 automatic correction)
AI disclosure Automatic metadata attachment (media-ai-disclosure chain)

Source: Direct quotation of PDF §6.8 ⑬ pass criteria

Related skills

  • Previous step ← media-moodboard (Deliverable ⑫ — color palette and tone keyword input)
  • Next step → moai-commerce:detail-page-image (Deliverable ⑭ — composite PNG, no changes to existing skills)
  • Automatic chain → media-ai-disclosure (automatic attachment of AI-generated labeling)
  • image-gen — single image generation, when multi-model routing is needed

When not to use this skill

Situation Skill to use instead
Single image generation (no set needed) image-gen
Moodboard and color palette establishment media-moodboard
Higgsfield Soul and fal.ai images image-gen or fal-gateway
Advertising video generation media-model-router
GPT-5.5 (LLM) or Sora (video) Not used in this course (REQ-MEDIA-014)