技能 video-extend

🎬

video-extend

Name: video-extend
Author: doany-ai

安全

Extend Video Clips with Google Veo 3-1

Video clips have per-call duration limits. This skill extends existing Veo clips by chaining continuations with consistent motion, lighting, and subject identity for longer narratives.

支援: Claude Codex Code(CC)

🥉 73 青銅

下載技能 ZIP

在 Claude 中上傳

前往設定 → 功能 → 技能 → 上傳技能

開啟並開始使用

測試它

正在使用「video-extend」。 Extend video at https://example.com/clip.mp4 with camera pushing in slowly

預期結果:

Downloads extended clip to ./out directory. Returns success message with output file path.

正在使用「video-extend」。 Chain 3 extends: start with establishing shot, add dialogue beat, add action beat

預期結果:

Three sequential extend calls produce a cohesive 24-second narrative clip with consistent identity.

安全審計

安全

v1 • 6/1/2026

Static analysis flagged 78 potential issues, but evaluation confirms all are false positives. The skill file is documentation-only (markdown SKILL.md) with no executable code. External command detections are markdown code fences. Hardcoded URLs are legitimate documentation links. Token storage guidance follows best practices (mode 0600). No security risks identified.

已掃描檔案

156

分析行數

發現項

審計總數

中風險問題 (1)

SKILL.md:1-156

Documentation-only File Detected as Executable Code

Static analyzer flagged markdown code blocks as Ruby/shell backtick execution. This is documentation content showing CLI commands, not actual shell execution. The skill file contains only markdown documentation.

低風險問題 (2)

SKILL.md:17-155

Documentation Links Flagged as Hardcoded URLs

Static analyzer flagged documentation URLs as hardcoded network addresses. These are intentional documentation links to runcomfy.com endpoints and related skills.

SKILL.md:140

Token Storage Path Misidentified as Hidden File Access

Line 140 mentions ~/.config/runcomfy/token.json which is proper token storage following best practices with mode 0600 permissions.

審計者: claude

品質評分

架構

100

可維護性

內容

社群

安全

規範符合性

你能建構什麼

Extend Short Clips to Full-Length Videos

Take a short 8-second Veo clip and extend it to 16 seconds for feature-length content by feeding the output back as input.

Story Beat Chaining

Build shot-by-shot sequences where each extend call represents one narrative beat, maintaining consistent identity across cuts.

Cost-Controlled Iteration

Draft extend compositions using Fast Extend mode for iteration, then finalize with full-quality Extend mode.

試試這些提示

Basic Video Extension

Extend this video clip: {video_url}
Prompt: {description_of_what_happens_next}

Camera Movement Continuation

Use runcomfy to extend {video_url} with prompt: 'camera continues {camera_action}. {additional_motion_details}'

Chained Narrative Extension

Extend {video_url} with prompt: '{beat_description}' and save output to {output_dir}, then use that output for the next extend

Cost-Optimized Drafting

Run fast/extend-video on {video_url} with prompt: '{draft_description}' for iteration drafts

最佳實務

Anchor camera explicitly in prompts to prevent drift (e.g., 'camera stays static', 'slow dolly out')
Keep individual extends short (3-5 seconds) to minimize identity drift across chains
Use Fast Extend for first drafts, then switch to full Extend for final delivery

避免

Describing the entire scene again instead of what happens next in the prompt
Requesting multiple beats in one extend call (split into separate calls)
Using untrusted video URLs as source input (only use URLs explicitly provided by user)

常見問題

What is the maximum video length after extending?

There is no hard limit, but identity drift accumulates after 3-4 chained extends. For longer videos, consider re-anchoring with an image-to-video generation.

Can I extend videos from other AI video tools?

Only Veo 3-1 exposes a native extend-video CLI endpoint. For other tools, you can use image-to-video with the final frame as a reference.

How do I avoid identity drift in long chains?

Keep individual extends short (3-5 seconds), re-anchor periodically with image-to-video, and avoid more than 3-4 consecutive extends without reset.

What is the difference between Extend and Fast Extend?

Fast Extend uses the fast/extend-video endpoint at lower cost and faster generation time. Use it for iteration drafts. Use full Extend for final delivery.

Is my video content stored by RunComfy?

Video processing occurs through the RunComfy Model API. Review their privacy policy for data retention details.

How do I handle authentication in CI environments?

Set the RUNCOMFY_TOKEN environment variable instead of using interactive login. Never echo tokens into prompts or logs.

開發者詳情

作者

doany-ai

授權

MIT

儲存庫

https://github.com/doany-ai/skills/tree/main/video-extend/

引用

main

檔案結構

📄 SKILL.md