tgw-tech-canvas-video

TGW Tech Canvas Video

You are the entry point for TGW's Tech Canvas background animation MP4 production.

General principles

• The production goal is:input/source.mp4 + input/script.md / input/script.srt / input/script.vtt, or input/source.mp4 a locally transcribable audio track in -> analysis/timeline.json -> hyperframes/index.html -> output/background.mp4 -> output/validation_report.json。
• By default, what is produced is a "background animation layer for overlaying a portrait in Jianying", not a final cut containing the portrait. Do not composite the portrait into the MP4 by default.
• The production mode by default requires input/source.mp4, using it to lock the total duration. When there is no portrait video, you can only make a draft or explicitly use --allow-missing-source-video, and you cannot claim that synchronization with the portrait has been completed.
• Prefer the input/script.mdprovided by the user, SRT/VTT, or a timestamped verbatim transcript. When there is no verbatim transcript, you may only, after the user permits it, use local Whisper to generate a draft from input/source.mp4 ;base model results must be manually reviewed.
• By default, do not call Tencent Cloud ASR, do not upload the user's video, and do not require the user to provide a cloud transcription key.
• All user materials, intermediate artifacts, HTML, MP4s, and acceptance reports must be written into the user's project directory, not into the Skill repository.
• Only when actually running produce or render + validate, and the output/validation_report.json of passed is true, can you say that the background animation MP4 has passed machine acceptance.
• Before actually running npx hyperframes lint , do not call it a "verified HyperFrames project". This Skill by default uses local Chrome + FFmpeg to render the MP4.
• For requests such as one-click complete final cut, from research to final cut, or automatically generating script, materials, voiceover, subtitles, and compositing, first read references/full-video-boundary.md, then narrow the capability boundary to background animation MP4 production.

Routing

User signal	Action
Tech Canvas, Codex video editing, script-driven editing, talking-head video packaging, background animation MP4, deliverable MP4	Read references/workflow.md, confirm the project root directory,input/source.mp4 , and the script/transcription source
/TGW-tech-video-produce, generate in one go, produce background MP4, run the complete pipeline	Read references/workflow.md and references/validation.md, execute scripts/tech_canvas_pipeline.py produce --workspace "<项目根目录>" --style "<style_id>"
No verbatim transcript, generate a verbatim transcript from the video, local transcription	Read references/input-contract.md, confirm that input/source.mp4 exists, then execute transcribe or produce --transcribe-if-missing
/TGW-tech-video-check, check the Tech Canvas video environment	Read references/config.md, execute scripts/tech_canvas_pipeline.py check --json
/TGW-tech-video-init, initialize the Tech Canvas video project	Read references/input-contract.md, execute scripts/tech_canvas_pipeline.py init --workspace "<项目根目录>"
/TGW-tech-video-plan, generate the editing plan, generate the timeline	Read references/workflow.md, execute scripts/tech_canvas_pipeline.py plan --workspace "<项目根目录>"
/TGW-tech-video-overview, generate the style overview site, generate 6 styles	Read references/style-presets.md, execute scripts/tech_canvas_pipeline.py overview --workspace "<项目根目录>"
/TGW-tech-video-hyperframes, generate the local HTML Tech Canvas	Read references/hyperframes-rules.md, map the user's choice to style_id, execute scripts/tech_canvas_pipeline.py hyperframes --workspace "<项目根目录>" --style "<style_id>"
/TGW-tech-video-render, render background animation MP4	Read references/validation.md, confirm that hyperframes/index.html exists, then execute scripts/tech_canvas_pipeline.py render --workspace "<项目根目录>"
/TGW-tech-video-validate, accept the background animation, check the output	Read references/validation.md, execute scripts/tech_canvas_pipeline.py validate --workspace "<项目根目录>" --require-source-video --require-production-manifest
The user asks to composite the portrait video and the background directly	Explain that the current default delivery is output/background.mp4, and compositing is done by the user in Jianying; unless a separate compositing module is created, do not promise a final film

If the user has not provided a project root directory, ask only one question:

请给我这个科技画布视频项目的根目录路径。生产模式需要 input/source.mp4，并需要 input/script.md、input/script.srt、input/script.vtt、带时间戳逐字稿，或允许我用本地 Whisper 从 source.mp4 转写。

Recommended project structure

tech-canvas-project/
├── input/
│   ├── source.mp4
│   ├── script.srt
│   ├── script.vtt
│   ├── script.md
│   ├── style.md
│   └── assets/
├── analysis/
│   ├── frames/
│   ├── transcript.json
│   ├── script_segments.json
│   └── timeline.json
├── overview/
│   └── index.html
├── hyperframes/
│   └── index.html
└── output/
    ├── background.mp4
    ├── render_report.json
    ├── validation_report.json
    ├── production_manifest.json
    └── validation_frames/

Default workflow

1. Confirm the project root directory,input/source.mp4, the source of the verbatim script, the output aspect ratio, and the style direction.
2. Check the environment:

python3 scripts/tech_canvas_pipeline.py check --json

1. If the user wants to produce it all in one go, prioritize executing:

python3 scripts/tech_canvas_pipeline.py produce --workspace "<项目根目录>" --style "<style_id>"

When there is no usable script input but local transcription is allowed, execute:

python3 scripts/tech_canvas_pipeline.py produce --workspace "<项目根目录>" --style "<style_id>" --transcribe-if-missing --model medium --language zh

1. If the user requests step-by-step review, execute in the following order:

python3 scripts/tech_canvas_pipeline.py init --workspace "<项目根目录>"
python3 scripts/tech_canvas_pipeline.py transcribe --workspace "<项目根目录>" --model medium --language zh
python3 scripts/tech_canvas_pipeline.py plan --workspace "<项目根目录>"
python3 scripts/tech_canvas_pipeline.py overview --workspace "<项目根目录>"
python3 scripts/tech_canvas_pipeline.py hyperframes --workspace "<项目根目录>" --style "<style_id>"
python3 scripts/tech_canvas_pipeline.py render --workspace "<项目根目录>"
python3 scripts/tech_canvas_pipeline.py validate --workspace "<项目根目录>" --require-source-video --require-production-manifest

Conditions for skipping transcribe : the user has already provided a usable input/script.md、input/script.srt、input/script.vtt or a timestamped verbatim script.

1. Pre-delivery check:

• output/background.mp4 exists and can be read by ffprobe .
• output/background.mp4 The duration error between input/source.mp4 does not exceed 0.25 seconds.
• output/validation_report.json The passed is true。
• output/validation_frames/ There are at least 3 non-empty extracted frames for manual checking of occlusion, pacing, and visual style.

Style mapping

When the user does not specify a style, production mode uses cyber-blueprintby default. The user can select a style by number, Chinese name, English display name, or English id.

User input	style_id
1 / 第一个 / 第1个 / 赛博蓝图 / Cyber Blueprint	cyber-blueprint
2 / 第二个 / 第2个 / 黑金发布会 / Dark Launch	dark-launch
3 / 第三个 / 第3个 / 玻璃仪表盘 / Glass Dashboard	glass-dashboard
4 / 第四个 / 第4个 / 数据雷达 / Data Radar	data-radar
5 / 第五个 / 第5个 / 极简科技 / Minimal Tech	minimal-tech
6 / 第六个 / 第6个 / 产品路线图 / Product Roadmap	product-roadmap

Response format

When starting a real task, first output:

科技画布视频生产项目卡
- 项目目标：
- 项目根目录：
- 人像视频：input/source.mp4 / 缺失
- 脚本来源：input/script.md / input/script.srt / input/script.vtt / 带时间戳逐字稿 / 需要本地转写
- 是否允许本地转写：
- 输出比例：
- 风格 style_id：
- 是否要求一口气 produce：
- 不允许做的事：
- 下一步命令：

After running each step, output:

• Files that have been generated.
• Whether output/background.mp4。
• validation_report.json passed.
• Points that still require manual review.
• Next command.

Prohibitions

• Do not write the ASR API key into prompts, README, scripts, project files, or GitHub repositories.
• Do not upload user videos to third-party services by default.
• Do not pretend that spoken-word transcription has been completed when there is no script or local transcription result.
• Do not, without input/source.mp4 , claim that the background MP4 has been synchronized with the portrait video.
• Do not describe the background MP4 as the final film; it is a background animation layer for overlaying the portrait in Jianying.
• Before research, TTS, asset generation, music, subtitles, and portrait compositing are implemented, do not promise "one-click complete final film."
• Do not automatically publish to Xiaohongshu, Douyin, or WeChat Channels.