viral-breakdown-to-video-prompts

name	viral-breakdown-to-video-prompts
description	NexTide 爆款拆解并反推视频提示词。调用小程序「爆款复刻 / 智能复刻」viral_clone 分镜拆解链路，适合“拆解这个爆款视频”“反推 Seedance/Veo 视频提示词”“把参考视频变成复刻分镜”。
allowed-tools	Read, Write, Bash

Viral Breakdown To Video Prompts

Follow shared NexTide rules in:

• nextide-shared

Use this skill when the user provides a viral video URL or uploaded reference video and wants a reusable breakdown / prompt base.

This skill is now aligned with the miniapp smart remix path:

小程序爆款复刻 → 智能复刻 → viral_clone storyboard breakdown

For image-text note/card references, use the image-text replication / Xiaohongshu skills instead.

Capability

Capability id:

viral.breakdown.video_prompts

CLI contract:

npm run nextide -- capability run viral.breakdown.video_prompts \
  --input .nextide/input/viral-breakdown-video-prompts.json \
  --output .nextide/output/viral-breakdown-video-prompts-result.json \
  --mode submit \
  --user-api-key <NEX用户积分API_KEY>

Mandatory Intake

Before submitting viral.breakdown.video_prompts, always confirm the following if the user has not already specified them:

1. Reference video: either an uploaded local video (upload to OSS first and pass the URL as referenceVideo) or a referenceUrl.
2. Target language: use Chinese labels in the interaction: 跟随原视频 / 中文 / 英文 / 日语 / 韩语 / 西语. Internally map them to source / zh-CN / en / ja / ko / es.
3. Prompt format: seedance / veo / generic.
4. Next step: breakdown only / continue generating video clips.

Do not ask for target duration or product binding in this intake. The backend keeps a compatibility duration for n8n, but the Agent UX should stay video + parameters only. Do not submit the smart remix job until at least one video reference and targetLanguage are confirmed.

Recommended interactive prompt:

请选择复刻参数：

参考视频：
A. 上传本地视频（自动上传 OSS）
B. 粘贴视频链接

目标语言：
A. 跟随原视频
B. 中文
C. 英文
D. 日语
E. 韩语
F. 西语

提示词格式：Seedance / Veo / 通用
下一步：只拆解提示词 / 继续生成视频片段

Input: Video Reference

{
  "referenceVideo": "https://.../video.mp4",
  "referenceUrl": "https://.../post",
  "sourcePlatform": "tiktok",
  "description": "可选参考说明",
  "targetLanguage": "zh-CN",
  "promptProvider": "seedance",
  "nextStep": "breakdown_only"
}

The runner calls the miniapp smart remix API:

POST /api/miniapp/storyboard/viral-clone/jobs

with pipeline_key=viral_clone. It returns a StoryboardTask id and usually enters waiting_callback until n8n calls back.

Output Handling

The callback writes structured results to StoryboardTask.detailedBreakdown and may also create StoryboardSegment rows for downstream image/video generation.

For Agent display, prefer the HTML report generated from:

StoryboardTask.detailedBreakdown.source_video_analysis
StoryboardTask.detailedBreakdown.content_structure
StoryboardTask.detailedBreakdown.beat_map
StoryboardTask.detailedBreakdown.full_original_script
StoryboardTask.detailedBreakdown.full_rewritten_script
StoryboardTask.detailedBreakdown.viral_mechanism
StoryboardTask.detailedBreakdown.clone_prompt.clips
StoryboardTask.detailedBreakdown.storyboard_grid_url

The primary prompt output is clone_prompt.clips[] (clip-level Seedance/Veo prompts), not each individual storyboard segment prompt.

Initial submit returns:

{
  "taskId": "...",
  "status": "ANALYZING"
}

After callback, run artifacts export and return the HTML report first:

nextide run artifacts <run-id> \
  --output-dir .nextide/output/<run-id> \
  --download \
  --gallery \
  --datatable

Return breakdown-report.html / preview.html first. The report must show the storyboard grid image, source video analysis, rhythm breakdown, original voiceover, rewritten voiceover, viral mechanism, and clip-level prompts.

After returning the report, ask whether the user wants product replacement:

是否需要继续替换产品？你可以上传产品图，我会直接调用 image2，默认替换所有出现产品的分镜，并在保持原分镜视觉结构完全不变的前提下，仅替换产品元素。若参考画面中有人物，我会避免使用真人脸作为参考，并要求脸部模糊/匿名化；Seedance 2.0 不支持真人图片参考。

If the user uploads a product image, use viral.breakdown.product_replace (Agent direct image2), not the old n8n storyboard image workflow.

Prompt Rule

This capability creates the smart-remix storyboard prompt base. To refine final Seedance/Veo prompts, combine with:

• reference-opening-decoder
• video-prompt-preflight-qa
• seedance

Reference Contract

Always separate:

• learn: hook structure, pacing, scene grammar, information hierarchy
• do not copy: person identity, brand, exact text, logo, creator face, proprietary visual elements

Rules

• Do not invent or default required intake values. If both referenceVideo and referenceUrl are missing, ask first. If targetLanguage is missing, ask first.
• For product replacement after the report, call Agent direct image2 capability viral.breakdown.product_replace; do not use the old image2/storyboard n8n workflow.
• Seedance 2.0 does not allow real-person image references. Do not accept or pass real human face images as references. If a reference frame includes a person, require face blur/anonymization and only replace products.
• Do not invent prompts if storyboard breakdown is pending.
• If only a platform post URL is provided and it is not a direct playable video URL, collect/normalize it first with the relevant collector skill.
• Treat this as a smart-remix storyboard breakdown capability, not final video generation.

NexTide Capability Contract

爆款拆解并反推视频提示词

• Capability: viral.breakdown.video_prompts
• Version: 0.2.0
• Category: video
• Status: available
• Execution: internal_api
• Async: true
• Cost level: medium
• Required auth: nexTideApiKey
• Required env: none
• Required plan: none
• Rate limit: 10/minute, 60/hour
• Estimated credits: 8
• Estimated duration: 180 seconds
• Tags: viral-breakdown, video-prompts, replication, viral-clone, smart-remix

Description:

调用小程序「爆款复刻 / 智能复刻」viral_clone 分镜拆解链路，拆解参考视频并生成 HTML 拆解报告、分镜网格图和可用于 Seedance/Veo 的 Clip 级分段提示词。

Examples:

• none

Input fields:

• referenceUrl (string)：参考爆款视频链接；没有 referenceVideo 时会作为 reference_video_url 使用。
• referenceVideo (string)：参考视频 URL，优先使用。
• targetLanguage (string, required)：目标语言。必须由用户用中文选项确认：跟随原视频 / 中文 / 英文 / 日语 / 韩语 / 西语；内部值分别映射为 source / zh-CN / en / ja / ko / es。
• promptProvider (string)：目标视频模型或提示词格式。 默认："seedance"
• nextStep (string)：breakdown_only 或 generate_clips。默认："breakdown_only"

Output fields:

• taskId (string)：StoryboardTask ID。
• status (string)：任务状态，例如 ANALYZING / BREAKDOWN_COMPLETED。
• breakdown (object)：回调完成后写入 StoryboardTask.detailedBreakdown 的爆款拆解结果。
• storyboardGridUrl (string)：分镜网格图 URL。
• clipPrompts (array)：detailedBreakdown.clone_prompt.clips 中的 Clip 级 Seedance/Veo 提示词。
• htmlReport (file)：Agent artifacts 导出的 breakdown-report.html。
• Follow-up product replacement: viral.breakdown.product_replace directly calls image2 from Agent. Inputs: productImage, referenceFrame or storyboardGridUrl, optional taskId/sceneOrder/prompt. If sceneOrder is not provided, default to replacing all scenes/segments where the product appears. Output: replaced product images, per-scene replacement results, and privacy notes.

CLI:

nextide capability run viral.breakdown.video_prompts \
  --input .nextide/input/viral.breakdown.video_prompts.json \
  --output .nextide/output/viral.breakdown.video_prompts-result.json \
  --mode submit \
  --wait \
  --timeout 900 \
  --interval 5

RUN_ID=$(node -e "const r=require('./.nextide/output/viral.breakdown.video_prompts-result.json'); console.log(r.run && r.run.runId)")
nextide run follow "$RUN_ID" \
  --output-dir .nextide/output/$RUN_ID \
  --timeout 900 \
  --interval 5

Artifact-first reading order:

1. Read .nextide/output/$RUN_ID/summary.json.
2. Read .nextide/output/$RUN_ID/manifest.json.
3. Return preview.html / gallery.html with rich preview when supported.
4. Return datatable.json for data/table results.
5. Return local artifact paths when present.
6. If a remote URL artifact is present, return the URL from manifest.
7. Only inspect the full result JSON when manifest is insufficient.

General Rules

• Use NexTide capability IDs, not internal n8n webhook URLs.
• Do not expose API secrets or internal webhook URLs in prompts or output.
• If status is not available, fail fast and explain what is missing.
• For async tasks, prefer --wait when the user wants a finished result in the same turn.
• After a finished run, use nextide run artifacts <run-id> --output-dir .nextide/output/<run-id> --download --gallery --datatable and read summary.json then manifest.json.
• For long-running runs, prefer nextide run follow <run-id> --output-dir .nextide/output/<run-id> --timeout 1800 --interval 5.
• Prefer returning summary.recommendedResponse.message, preview.html, datatable.json, and local artifact paths over pasting huge raw JSON.
• If the CLI output includes explanation, convert it into a clear user-facing failure message with next actions.