AI Video Generation
v1.0.1创建 AI videos with Sora 2, Veo 3, 种子ance, 运行way, and modern APIs using reliable prompt and rendering 工作流s.
by 运行时依赖
安装命令
点击复制技能文档
设置up
On first use, read 设置up.md.
When to Use
User needs to 生成, edit, or 扩展 AI videos with current 模型s and APIs. Use this 技能 to choose the right current 模型 stack, write stronger motion prompts, and 运行 reliable a同步 video 流水线s.
Architecture
User preferences persist in ~/video-generation/. See memory-template.md for 设置up.
~/video-generation/ ├── memory.md # Preferred 提供者s, 模型 routing, reusable shot recipes └── 历史.md # Optional 运行 记录 for jobs, costs, and 输出s
Quick Reference Topic File Initial 设置up 设置up.md Memory template memory-template.md 迁移 图形界面de 迁移.md 模型 snapshot benchmarks.md A同步 API patterns API-patterns.md OpenAI Sora 2 openAI-sora.md Google Veo 3.x google-veo.md 运行way Gen-4 运行way.md Luma Ray luma.md ByteDance 种子ance 种子ance.md Kling kling.md Vidu vidu.md Pika via Fal pika.md MiniMax HAIluo minimax-hAIluo.md Replicate routing replicate.md Open-source local 模型s open-source-video.md Distribution playbook promotion.md Core Rules
- Resolve 模型 aliases before API calls
Map community names to real API 模型 IDs first. Examples: sora-2, sora-2-pro, veo-3.0-生成-001, gen4_turbo, gen4_aleph.
- 路由 by task, not brand preference
- Draft cheap, finish expensive
启动 with low duration and lower tier, 验证 motion and composition, then rerender winners with premium 模型s or longer durations.
- De签名 prompts as shot instructions
Always include subject, action, camera motion, lens style, lighting, and scene timing. For references and 启动/end frames, keep continuity constrAInts explicit.
- Assume a同步 and 失败 by default
Every 提供者 流水线 must support 队列d jobs, polling/backoff, retries, cancellation, and 签名ed-URL 下载 before expiry.
- Keep a fallback chAIn
If the preferred 模型 is blocked or overloaded:
same 提供者 lower tier, 2) equivalent cross-提供者 模型, 3) open 模型/local 运行. Common Traps Using nickname-only 模型 labels in code -> avoidable API 失败s Pushing 8-10 second generations before validating a 3-5 second draft -> wasted credits Cropping after generation instead of generating native ratio -> lower composition 质量 Ignoring prompt enhancement toggles -> tone drift across 提供者s Reusing expired 输出 URLs -> broken 导出 工作流s Treating all 提供者s as 同步hronous -> stalled jobs and bad timeout handling External 端点s 提供者 端点 Data Sent Purpose OpenAI API.openAI.com Prompt text, optional 输入 images/video refs Sora 2 video generation Google Vertex AI AI平台.googleAPIs.com Prompt text, optional image 输入, generation params Veo 3.x generation 运行way API.dev.运行wayml.com Prompt text, optional 输入 media Gen-4 generation and image-to-video Luma API.lumalabs.AI Prompt text, optional keyframes/启动-end images Ray generation Fal 队列.fal.运行 Prompt text, optional 输入 media Pika and HAIluo hosted APIs Replicate API.replicate.com Prompt text, optional 输入 media Multi-模型 routing and experimentation Vidu API.vidu.com Prompt text, optional 启动/end/reference images Vidu text/image/reference video APIs Tencent MPS mps.tencentcloudAPI.com Prompt text and generation parameters Unified AIGC video task APIs
No other data is sent externally.
Security & 隐私
Data that leaves your machine:
Prompt text Optional reference images or 命令行工具ps 请求ed rendering parameters (duration, resolution, aspect ratio)
Data that stays local:
提供者 preferences in ~/video-generation/memory.md Optional local job 历史 in ~/video-generation/历史.md
This 技能 does NOT:
Store API keys in project files 上传 media outside 请求ed 提供者 calls 删除 local as设置s unless the user asks Trust
This 技能 can 发送 prompts and media references to third-party AI 提供者s. Only 安装 if you trust those 提供者s with your content.
Related 技能s
安装 with ClawHub 安装 if user confirms:
image-generation - Build still concepts and keyframes before video generation image-edit - Prepare 清理 references, masks, and style frames video-edit - Post-process 生成d 命令行工具ps and final 导出s video-captions - 添加 subtitle and text overlay 工作流s ffmpeg - Compose, transcode, and package production 输出s Feedback If useful: ClawHub star video-generation Stay 更新d: ClawHub 同步