Real Human + Named Anime IP Photo Template

date: '2026-05-01T14:03:30+08:00'
result: pass
prompt:
  text: >-
    Create a highly realistic photo where I am standing next to Ninja Hattori and his friend Shinzo
    Hattori. Both characters should appear slightly stylized but naturally blended into a real-world
    environment. Ninja Hattori should be in his classic ninja outfit with a calm and alert
    expression, and Shinzo should look small, cute, and playful. I am standing confidently beside
    them, posing for a photo. Keep my face same 100% as in the reference image. The background
    should be a simple outdoor setting like a street or park with natural lighting. My outfit should
    be stylish and modern ("zabardast"), and the overall image should look like a real-life
    photograph with high detail and natural expressions.
refs:
  - source: yuan.studio/public/images/yuan_kid_history/raw_frame_001.png
    sha256: null
    error: file not found at write time
    upload_url: >-
      https://personalized-video-refs.tos-cn-beijing.volces.com/4a21e4ba8a91/raw_frame_001.png?X-Tos-Algorithm=TOS4-HMAC-SHA256&X-Tos-Credential=AKLTYjM3ZTcxNjI5ZjFkNGZmYjgyMTNjNWRjZjU4N2IzNjY%2F20260430%2Fcn-beijing%2Ftos%2Frequest&X-Tos-Date=20260430T175741Z&X-Tos-Expires=86400&X-Tos-SignedHeaders=host&X-Tos-Signature=e00dec5d3246089dde91813f6c7914ef6c2b1d22eaca22aff4207093249a22d0
provider:
  id: gpt_image_2
  relay: apimart
  config:
    aspect_ratio: '1:1'
    size: '1:1'
    'n': 1
output:
  path: ./anime_ip_blend_yuan_kid_v1.png
  bytes: 2311897
  wall_seconds: 40.3
  task_id: task_01KQH240XQS5NXMBXPCR7W4XE8
script: experiments/anime_ip_blend_test/test_v1_yuan_kid.py
cost_yuan: 0.5
notes: |
  7 unique 维度 · 6 PASSED + 1 partial · 40s wall.

  **3 个 framework-级新发现:**

  (a) **gpt-image-2 RECOGNIZES + REPRODUCES specific anime IP** (Ninja Hattori)
      · 不是 filter · 不是 generic ninja 替换 · 出片清晰 Hattori (蓝服 + 白头巾 + 背刀 + 圆脸)
      · "Shinzo" 被解为 K-cap 角色 (像 Kenichi Mitsuba style · 模型 IP 知识有局限)
      · ApiMart endpoint 确实 image gen 政策宽松 (memory feedback_apimart_endpoints_audit_diff.md 验证 +1)
      · **production 含义**: 个人 + 经典动漫 IP 同框可做 · 但 IP 准确性有偏差 · 商用须法务

  (b) **CLEVER 解 photoreal-stylized 冲突** · 渲染为"主题乐园 mascot 表演者"
      · Anime characters 看似 walk-around mascot suit 中的真人
      · 解决"同框 photoreal 真人 + cartoon"难题 · 不是简单 cel-shading overlay
      · 模型自创 framework: "stylized" → "costumed performer at theme park"
      · 启示: anchor 冲突时 model 偏向找 real-world plausible 解

  (c) **3-character composition 稳定**
      · yuan_kid 中央 + 双侧 anime · triangle 构图
      · 比例正确 (Hattori 大人尺寸 · K-cap 角色小 + 矮一点)
      · 全员 peace sign / posing for camera · pose coherent
      · 之前我们没测过 user + 2 character · 现在 validate

  其他维度:
    · 100% face lock: yuan_kid 清晰可辨 · 轻微 age compensation (~5yo vs 4yo)
    · "Stylish modern outfit": 粉色 jacket + 白底 graphic tee + 卡其裤 · 4yo-appropriate stylish
    · "Zabardast" (Hindi/Urdu 'fantastic') 大概率被忽略 · outfit 不是 fantastically standout
    · Outdoor street + natural lighting + tree/buildings background ✓
    · Selfie meta-pose ('confidently posing') ✓

  Production use:
    · 个人 + 经典动漫 IP 合影 · 粉丝周边 / SNS 内容
    · 主题乐园 mascot 互动模拟
    · 跨 style "我和 X" 系列 (X 可换任何 cartoon IP)
    · ApiMart 通道证实可用 · 但 IP 精确度待 case-by-case 验

  GOTCHA · 需 case-by-case 验 IP 是否被 model 正确识别 (Hattori 准 · Shinzo 不准 · Pikachu/Mickey 等更出名 IP 待测).
  GOTCHA · 多语言 fashion vocab (Hindi/Urdu/etc.) 大概率被忽略 · 用纯 EN/CN 描述.