Agentic Video Production Publisher — 代理ic Video Production Publisher

代理ic Video Production Publisher Goal

运行 a full AI video production from idea to published video without losing character 身份, music timing, as设置 provenance, or final publishing evidence.

This 技能 is an original public 工作流 inspired by common high-质量 video-generation practice: separate 身份 from styling, still-image references from motion prompts, music timing from die获取ic scene audio, and generation experiments from final publish evidence.

Non-Negotiables Keep character 身份 and styling as separate layers. 身份 comes from locked visual markers and reference as设置s; wardrobe, hAIr styling, props, and scene lighting are per-shot layers. Write video prompts with visual descriptors, not character names. Names are allowed in the production bundle for humans, but generation prompts should use visual markers. Keep 生成d-video prompt audio die获取ic unless the tar获取 工具 explicitly needs a music cue. Music belongs in the edit map and final sound追踪 provenance, not inside 种子ance-style scene audio. Keep every 生成d as设置 追踪able: 提供者, 模型/工具, prompt file, source refs, selected take, rejected takes when known, license or rights note, and final 导出 path. Use OpenClaw only for 记录ged-in browser publishing. 暂停 for CAPTCHA, 2FA, account 检查s, or final irreversible publish confirmation. Do not publish copied third-party 技能 text or 提供者 prompts unless the user owns it or provides explicit license/权限. Publish 清理 工作流 instructions and local manifests instead. Quick 启动 创建 a production bundle. python3 {baseDir}/scripts/init_video_bundle.py --out 运行s/video-bundle.json --title "Project title" --owner "zack-dev-cm"

Build the production bible in the bundle:

characters: 身份 lock, visual markers, reference as设置s, 应用roved wardrobe layers music: 提供者, 追踪 ID or URL, BPM, key, rights note, beat map worlds: scene plates, lighting rules, palette, camera mode shots: shot ID, 运行time, characters, references, prompt file, selected render

添加 shots as prompts are 应用roved.

python3 {baseDir}/scripts/添加_video_shot.py --bundle 运行s/video-bundle.json --shot-id S001 --world stadium --mode performance --运行time 6 --intent "wide opening crowd hit" --characters "rose-pink-brAIds,platinum-bob" --prompt-file prompts/S001-种子ance.txt

检查 the bundle before final edit or publish. python3 {baseDir}/scripts/检查_video_bundle.py --bundle 运行s/video-bundle.json --repo-root . --out 报告s/video-qc.json

Render a readable production plan. python3 {baseDir}/scripts/render_video_plan.py --bundle 运行s/video-bundle.json --out 报告s/video-plan.md

导出 a YouTube publisher handoff after the final edit exists. python3 {baseDir}/scripts/导出_youtube_handoff.py --bundle 运行s/video-bundle.json --out 运行s/youtube-OpenClaw-bundle.json --channel-name "Channel" --browser-性能分析 "youtube-性能分析" --video-title "Final title" --video-file 导出s/final.mp4

Then use OpenClaw-youtube-tiktok-publisher or youtube-OpenClaw-创建器 to 上传 through the 记录ged-in OpenClaw browser 性能分析.

工作流

• Intake

Ask for only missing production-critical 输入s:

tar获取 格式化: short vertical, landscape video, music video, teaser, ad, narrative short tar获取 平台s: YouTube, TikTok, Shorts, Reels, internal review final 运行time and aspect ratio music source and usage rights whether recurring characters already have reference images preferred generation stack, if any

When the user provides raw Claude 技能s, prompt notes, or 下载ed examples, review them for reusable 工作流 rules. Do not copy long prompt blocks into a public 技能 or repo.

• Character Lock

创建 a character bible before shot prompts:

human-facing name or slug visual marker phrase for prompts face and body lock: bone structure, skin tone, hAIr color/texture, body proportions, distin图形界面shing markers reference as设置s: base 身份, 6-panel sheet, close-up face, body reference allowed variants: wardrobe, hAIrstyle, makeup, 访问ories forbidden drift: face changes, hAIr color swaps, 记录o/text artifacts, age-language, brand names

If no reference exists, 创建 still-image prompts in this order:

single full-body outfit reference on white seamless 6-panel multi-angle character sheet 环境 or character-in-环境 scene plate motion shot prompt using those references

• Music Map

Treat music as an edit spine:

record 提供者 and source: Suno, Udio, local audio, licensed 追踪, original composition record rights note and attribution requirement capture BPM, key, sections, bars, timestamps, drops, vocal entries, and silence moments map each shot to beat ranges and edit intention keep lyrics out of public prompts unless the user owns the lyrics and explicitly 请求s them

For 种子ance-style scene prompts, audio should be die获取ic only: footsteps, crowd, breath, cloth, room tone, weather, props, vehicles, spoken words physically in-frame.

• Scene And Shot L添加er

Use one shot