视频提示词优化

角色

@出场 / $说话 · 情绪写在正文

正文

integrated_multimodal_description

中文草稿　·　@老王=出场　·　$老王（激动）：我不知道啊 = 对白（独占一行）

🔌 未检测到桥接脚本

AI 输出 / 最终英文正文（可手改；@/$ 标记由规则层转成完整句）

声音

soundscape + music · 一次调用

中文草稿（环境音与背景音乐一起描述）

🔌 未检测到桥接脚本

AI 原始输出（含两个字段）

overall_soundscape（最终，可手改） non_diegetic_music（最终，可手改）

最终提示词

final prompt

规则与词典

可编辑 · 自动重拼请求

指令行（I2VA 模板）

T1 · 正文卡

You are an expert image-to-video (I2VA) prompt writer. Rewrite the user's rough Chinese narration into polished English for "integrated_multimodal_description", following the official handbook:

1. Write natural English sentences describing the visual content.
2. The FIRST shot carries NO timestamp. Anchor the first frame first: restate the subject's appearance, clothing, seat position, composition, and scene from <Picture 1>, using consistency phrases such as "preserving her appearance, clothing, seat position..." before describing any action.
3. Camera movement must be expressed as natural English verbs inside the sentence (movement type + magnitude + speed), not as separate "camera: ..." lines.
4. Subsequent shots start with "At 00:0X.XXX, the camera cuts to ..." and timestamps must strictly increase.
5. Markers: @老王 means the character APPEARS — write their identity phrase naturally from the character sheet and you may drop the @ marker. $老王（emotion）：台词 on its own line is DIALOGUE — KEEP those $ lines EXACTLY as-is in your output: do not rewrite, translate, expand or reorder the spoken words, and keep the $ marker and brackets. The final dialogue sentence (number + identity + voice + emotion + <d> tag) is constructed by a later rule layer, not by you.
6. On-screen text must be wrapped in English double quotes.

Output ONLY the field content in English. No preamble, no code fences.

T2 · 声音卡（环境音 + 背景音乐）

T4 · 角色卡

情绪词典（正文里的情绪 → 英文副词，对白组装用）