MartinDelophy/ai-video-editor v1.0.9

Timeline Studio v1.0.9 — Agent relighting and depth video

on GitHub

Timeline Studio agents can now run depth-assisted relighting analysis and depth-video conversion through the browser's structured WebMCP tools, then review editable lighting changes or insert generated assets into the timeline.

Highlights

• Add relight-analysis and depth-map to the existing AI prepare → start → inspect → cancel lifecycle. Tasks bind to inspected Visuals/Overlay clip IDs and reuse the production depth worker, pinned owned model mirrors, renderer and RIFE interpolation.
• Add shared clip.set_relight commands with validated lighting parameters, semantic before/after review, track-lock protection and undo. Enabling lighting requires complete analysis matching the source timing and selected quality.
• Keep Agent-generated depth videos in My assets until a separate reviewed insertion; analysis does not silently enable effects or replace source clips. Source changes are checked before committing results.
• Report actual model/task progress, support real cancellation, and return generated asset receipts. Add direct translations for the new task/review copy in all 13 interface languages.
• Update the repository editing Skill, self-contained browser discovery Skill, discovery digest, integration docs and localized project updates.

This release also includes main's accumulated improvements since v1.0.8, including long-video playback, scrubbing and export optimizations. See the full changelog for the complete scope.

Execution boundaries

WebMCP requires a compatible native browser API and agent host. Depth inference requires WebGPU. Model execution and rendered lighting use the live browser pipeline; local CLI/STDIO MCP can diff/apply lighting settings but does not execute these model jobs. Headless rendering explicitly rejects enabled depth/lighting effects rather than silently omitting them. Browser-session analysis may require reanalysis after reopening a portable project.

Validation

• npm run check: lint, TypeScript and production build; 79 existing lint warnings remain.
• Real Chrome/WebGPU analysis and depth-video generation using the production worker and RIFE route; generated 320×240 depth output contains 15 frames at a nominal 24 fps and an Opus audio track.
• Rendered samples confirmed a visible lighting change; deterministic browser export produced a decodable H.264/AAC MP4.
• Verified semantic review, apply/undo, shared parameter validation, track locks, retry-key deduplication, cancellation and stale-source handling. Real worker cancellation acknowledged a cancelled terminal state.
• Saved generated media in a portable .timeline archive; local CLI diff/apply persisted relighting settings and the headless renderer rejected unsupported lighting.

中文说明

• 视频打光和深度视频转换正式接入 WebMCP 的准备、启动、进度查询和取消流程，复用现有生产管线。
• 打光分析只准备深度数据；参数通过 clip.set_relight 单独审阅、应用和撤销，启用前检查源片段与分析质量是否匹配。
• Agent 生成的深度视频先进入“我的资产”，随后通过独立审阅插入时间线，不自动替换原片段。
• 同步更新编辑 Skill、网站浏览器 Skill、发现索引摘要、13 种语言文案和相关文档。
• 本地 CLI/STDIO MCP 支持打光参数编辑；模型分析、转换及打光成片仍由浏览器执行。

Install the versioned editing Skill for Codex:

gh skill install MartinDelophy/ai-video-editor edit-timeline-studio --pin v1.0.9 --agent codex --scope user

Editing workflow · WebMCP integration · Full changelog