Personalized Korean baseball broadcast

스튜디오 초안을 준비합니다. 여러 단계의 지침은 참고용이며 자동 생성이나 과금은 하지 않습니다.

上传一张个人参考照片并提供一个直播名称，即可一键生成 30 秒真实的韩国职业棒球联赛（KBO）电视直播风格的场边摄像机镜头。

제작 흐름 본문은 중국어 원문입니다. 언급된 모델과 이 사이트의 지원 모델은 다를 수 있습니다. 실제 모델과 크레딧은 스튜디오에서 확인하세요.

전체 제작 흐름

1. 제작 기획

**前置检查（必须完成，否则不继续任何生产步骤）：**

开始前确认用户已提供以下两项：
1. **本人参考图**：用于保留面部特征的自拍照或清晰头像。
2. **播报称谓**：将出现在转播名字条上的文字（例如："VaVa - Flova AI Creator"）。
若任一缺失，立即告知用户并等待补充，不得跳过或使用默认值。

**生产流程：**
1. 以用户上传的参考图为基础，生成仿真 NBA 转播定格截图 → **media_generator**（GPT Image 2）
   - 完成后暂停，展示截图，等待用户确认人物还原效果。
2. 以步骤 1 截图为 start_frame、用户原始参考图为 reference_image，生成 30s 转播视频 → **media_generator**（Seedance 2.5（分辨率 480p））
3. 展示最终视频，提供确认或重新生成选项。

**依赖关系：** 2 → 1；（截图确认通过后方可继续）。
**暂停节点：** 步骤 1 完成后必须暂停，用户确认身份还原满意后再推进至视频生成。

前置检查（必须完成，否则不继续任何生产步骤）：

开始前确认用户已提供以下两项：

本人参考图：用于保留面部特征的自拍照或清晰头像。
播报称谓：将出现在转播名字条上的文字（例如："VaVa - Flova AI Creator"）。
若任一缺失，立即告知用户并等待补充，不得跳过或使用默认值。

生产流程：

以用户上传的参考图为基础，生成仿真 NBA 转播定格截图 → media_generator（GPT Image 2）
- 完成后暂停，展示截图，等待用户确认人物还原效果。
以步骤 1 截图为 start_frame、用户原始参考图为 reference_image，生成 30s 转播视频 → media_generator（Seedance 2.5（分辨率 480p））
展示最终视频，提供确认或重新生成选项。

依赖关系： 2 → 1；（截图确认通过后方可继续）。
暂停节点： 步骤 1 完成后必须暂停，用户确认身份还原满意后再推进至视频生成。

2. 미디어 생성

**步骤 1 — 生成直播静态截图**
- 工具：**ImageToImage**；模型：**GPT Image 2**；分辨率：**2K**。
- 参考图像：用户上传的个人参考照片（绑定至角色元素）。
- 生成后，注册 asset_id 并将图像绑定至镜头 1 的 start_frame 插槽。
- 只有在用户确认对截图中的角色相似度满意后，方可进行步骤 2。

**步骤 2 — 生成 30 秒场边直播视频**
- 工具：**MultiModalToVideo**；模型：**Seedance 2.5**（分辨率 **480p**）；分辨率：**480p**；时长：**30s**。
- 参考输入：
  1. **start_frame**：步骤 1 生成的直播截图（视觉起始帧参考）。
  2. **reference_image**：用户上传的原始参考照片（身份保留参考）。

步骤 1 — 生成直播静态截图

工具：ImageToImage；模型：GPT Image 2；分辨率：2K。
参考图像：用户上传的个人参考照片（绑定至角色元素）。
生成后，注册 asset_id 并将图像绑定至镜头 1 的 start_frame 插槽。
只有在用户确认对截图中的角色相似度满意后，方可进行步骤 2。

步骤 2 — 生成 30 秒场边直播视频

工具：MultiModalToVideo；模型：Seedance 2.5（分辨率 480p）；分辨率：480p；时长：30s。
参考输入：
1. start_frame：步骤 1 生成的直播截图（视觉起始帧参考）。
2. reference_image：用户上传的原始参考照片（身份保留参考）。

3. 프롬프트 작성

**全局规则：** 所有生成 prompt 以英文撰写；将用户提供的播报称谓直接替换模板中所有 \[NAME\] 占位符，不得使用默认值或虚构称谓。

### 图像 Prompt（ImageToImage：GPT Image 2）
参考图输入：<<<image_1>>> 绑定为用户本人参考图。
> <<<image_1>>> — A screenshot from a live KBO (Korea Baseball Organization) game TV broadcast on MBC Sports+. The camera cuts to the audience — the reference image person is one of the audience sitting between, sitting and smiling naturally, unaware they're on camera. The subject is sitting in the rows of the stands at Jamsil Baseball Stadium in Seoul, South Korea. **Hardlock: do not alter their facial structure; strictly preserve their likeness from the reference image.** Full KBO broadcast overlay: static scorebug at the bottom showing LG Twins vs Doosan Bears regular season matchup, LG Twins leading 3–1 in the 7th inning; MBC Sports+ network logo watermark in the top-left corner; 16:9 aspect ratio. Above the scorebug, a clean broadcast name graphic in Korean broadcast style reads: **"\[NAME\]"**, white text on a semi-transparent dark bar, styled like a real KBO broadcast identifier. The image looks exactly like a real Korean TV sports broadcast screenshot — broadcast color grading, slight compression artifacts, interlacing grain, Korean UI typography. All on-screen graphics are completely static, no animation.

### 视频 Prompt（MultiModalToVideo + Seedance 2.5（分辨率 480p））
参考图输入：<<<image_1>>> 绑定步骤 1 生成的转播截图（start_frame）；<<<image_2>>> 绑定用户原始参考图（identity reference）。
> <<<image_1>>> <<<image_2>>> — A realistic live KBO broadcast shot of the subject sitting in the front stands at Jamsil Baseball Stadium, Seoul, during an LG Twins vs Doosan Bears regular season game. The shot feels like a real TV cutaway when the MBC Sports+ broadcast camera finds a notable guest in the crowd. Preserve the subject's identity and likeness from the reference images throughout the entire clip.
>
> The subject is seated in the stadium stands, smiling naturally and not over-performing for the camera. Not locked into eye contact with the lens — they occasionally glance toward the field, then toward the camera, then back toward the field.
>
> One continuous take. No cuts. No angle changes.
>
> Action sequence: They smile casually in their seat as the camera lands on them, looking around naturally. They then give a relaxed, natural wave toward the camera — the crowd around them cheers and reacts. They glance up toward the scoreboard above the outfield wall, then look back at the camera. They cheer briefly with visible excitement reacting to the on-field action, then turn to say something to the person beside them and laugh — we don't hear them speak. They finish by clapping naturally while smiling. All movement is subtle, believable, and human. No exaggerated acting. No direct talking to camera.
>
> Broadcast style: real live Korean sports TV broadcast look, telephoto broadcast camera feel, natural outdoor stadium lighting under evening lights, slight broadcast compression, slight interlacing / TV grain, authentic crowd movement in background, Korean baseball stadium atmosphere with LG Twins and Doosan Bears fan sections visible. Subject remains in their stadium seat for the full 30 seconds.
>
> On-screen graphics: static bottom scorebug showing LG Twins vs Doosan Bears KBO regular season layout — completely unchanged for the entire 30 seconds, no animation, no score updates, LG Twins leading 3–1 in the 7th inning. MBC Sports+ logo watermark top-left corner. Above the scorebug, broadcast name graphic reads: **"\[NAME\]"** — Korean broadcast style, white text on semi-transparent dark bar, present for the full duration.
>
> {Two male KBO broadcast commentators in Korean, casual and warm tone: "\[NAME\] 씨가 오늘 잠실 구장을 찾아주셨습니다. 7회 접전이 펼쳐지는 가운데 관중석에서 LG와 두산의 명승부를 직접 지켜보고 계시네요. 팬들의 뜨거운 응원에 \[NAME\] 씨도 함께 즐기고 계신 것 같습니다."} <natural baseball stadium crowd noise, vendor sounds, and ambient stadium atmosphere throughout>
>
> no music, no subtitles, no scene cuts, no angle changes, no scorebug animation or updates, no exaggerated gestures, no constant direct eye contact with camera, no talking to camera

全局规则： 所有生成 prompt 以英文撰写；将用户提供的播报称谓直接替换模板中所有 [NAME] 占位符，不得使用默认值或虚构称谓。

图像 Prompt（ImageToImage：GPT Image 2）

参考图输入：<<<image_1>>> 绑定为用户本人参考图。

<<<image_1>>> — A screenshot from a live KBO (Korea Baseball Organization) game TV broadcast on MBC Sports+. The camera cuts to the audience — the reference image person is one of the audience sitting between, sitting and smiling naturally, unaware they're on camera. The subject is sitting in the rows of the stands at Jamsil Baseball Stadium in Seoul, South Korea. Hardlock: do not alter their facial structure; strictly preserve their likeness from the reference image. Full KBO broadcast overlay: static scorebug at the bottom showing LG Twins vs Doosan Bears regular season matchup, LG Twins leading 3–1 in the 7th inning; MBC Sports+ network logo watermark in the top-left corner; 16:9 aspect ratio. Above the scorebug, a clean broadcast name graphic in Korean broadcast style reads: "[NAME]", white text on a semi-transparent dark bar, styled like a real KBO broadcast identifier. The image looks exactly like a real Korean TV sports broadcast screenshot — broadcast color grading, slight compression artifacts, interlacing grain, Korean UI typography. All on-screen graphics are completely static, no animation.

视频 Prompt（MultiModalToVideo + Seedance 2.5（分辨率 480p））

参考图输入：<<<image_1>>> 绑定步骤 1 生成的转播截图（start_frame）；<<<image_2>>> 绑定用户原始参考图（identity reference）。

<<<image_1>>> <<<image_2>>> — A realistic live KBO broadcast shot of the subject sitting in the front stands at Jamsil Baseball Stadium, Seoul, during an LG Twins vs Doosan Bears regular season game. The shot feels like a real TV cutaway when the MBC Sports+ broadcast camera finds a notable guest in the crowd. Preserve the subject's identity and likeness from the reference images throughout the entire clip.

The subject is seated in the stadium stands, smiling naturally and not over-performing for the camera. Not locked into eye contact with the lens — they occasionally glance toward the field, then toward the camera, then back toward the field.

One continuous take. No cuts. No angle changes.

Action sequence: They smile casually in their seat as the camera lands on them, looking around naturally. They then give a relaxed, natural wave toward the camera — the crowd around them cheers and reacts. They glance up toward the scoreboard above the outfield wall, then look back at the camera. They cheer briefly with visible excitement reacting to the on-field action, then turn to say something to the person beside them and laugh — we don't hear them speak. They finish by clapping naturally while smiling. All movement is subtle, believable, and human. No exaggerated acting. No direct talking to camera.

Broadcast style: real live Korean sports TV broadcast look, telephoto broadcast camera feel, natural outdoor stadium lighting under evening lights, slight broadcast compression, slight interlacing / TV grain, authentic crowd movement in background, Korean baseball stadium atmosphere with LG Twins and Doosan Bears fan sections visible. Subject remains in their stadium seat for the full 30 seconds.

On-screen graphics: static bottom scorebug showing LG Twins vs Doosan Bears KBO regular season layout — completely unchanged for the entire 30 seconds, no animation, no score updates, LG Twins leading 3–1 in the 7th inning. MBC Sports+ logo watermark top-left corner. Above the scorebug, broadcast name graphic reads: "[NAME]" — Korean broadcast style, white text on semi-transparent dark bar, present for the full duration.

{Two male KBO broadcast commentators in Korean, casual and warm tone: "[NAME] 씨가 오늘 잠실 구장을 찾아주셨습니다. 7회 접전이 펼쳐지는 가운데 관중석에서 LG와 두산의 명승부를 직접 지켜보고 계시네요. 팬들의 뜨거운 응원에 [NAME] 씨도 함께 즐기고 계신 것 같습니다."} <natural baseball stadium crowd noise, vendor sounds, and ambient stadium atmosphere throughout>

no music, no subtitles, no scene cuts, no angle changes, no scorebug animation or updates, no exaggerated gestures, no constant direct eye contact with camera, no talking to camera