Hubbestaiskill.com

Home / All Skills / Content & Media Creation / ai-video-generation

Content & Media Creation · 101-skills/superpowers

ai-video-generation

Generate AI videos with Google Veo, Seedance 2.0, HappyHorse, Wan, Grok and 40+ models via inference.sh CLI. Models: Veo 3.1, Veo 3, Seedance 2.0, HappyHorse 1.0, Wan 2.5, Grok Imagine Video, OmniHuman, Fabric, HunyuanVideo. Capabilities: text-to-video, image-to-video, reference-to-video, video editing, lipsync, avatar animation, video upscaling, foley sound. Use for: social media videos, marketing content, explainer videos, product demos, AI avatars. Triggers: video generation, ai video, text to video, image to video, veo, animate image, video from image, ai animation, video generator, generate video, t2v, i2v, ai video maker, create video with ai, runway alternative, pika alternative, sora alternative, kling alternative, seedance, happyhorse

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandNetwork AccessDual-pass Audit

Author 101-skillsGitHub 101-skills/superpowers ↗License MIT（仓库 README 徽标 'License: MIT' 与 .claude-plugin/plugin.json 中 "license": "MIT" 声明；但 pin 树内无 LICENSE 文件，GitHub API 对两 fork 亦报无 license —— 授权只出现在元数据/徽标，无正式许可文本）Commit becc256497

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Isomorphic mechanism to twitter-automation: The sole execution interface is the belt CLI (allowed-tools: Bash(belt *)). The documentation structures '40+ models' into app-id directory tables grouped by purpose (Text-to-Video / Image-to-Video / Avatar-Lipsync / Editing / Utilities), invoked by the agent via --input JSON.

tools/video/ai-video-generation/SKILL.md

belt app run google/veo-3-1-fast --input '{"prompt": "drone shot flying over a forest"}'

Quick Start boils down to 'copy command + fill in prompt'; switching models or tasks only requires changing app-id and JSON keys, while model inference fully takes place on the inference.sh platform.

Model selection knowledge for 'which model fits which task' is directly tabulated: Veo 3.1 (quality/interpolation), Seedance 2.0 (1080p+synced audio), Wan 2.5 (i2v), HappyHorse (photorealism/15s), Grok (xAI), etc., each have a 'Best For' one-liner so the agent does not need to compare them autonomously.

tools/video/ai-video-generation/SKILL.md

| Veo 3.1 Fast | `google/veo-3-1-fast` | Fast, with optional audio |

The 'Best For' column serves as the model routing decision layer; selection rationale is preset by the author rather than inferred dynamically by the model.

The input contract relies entirely on remote URLs: image_url / audio_url / video_url must be provided by the user or generated by upstream apps; the skill contains zero 'local upload / read local file' logic—media assets must have a public URL before being passed to models.

tools/video/ai-video-generation/SKILL.md

"prompt": "A timelapse of a flower blooming in a garden"

This requires assets to be hosted on URL-accessible endpoints (cloud.inference.sh or elsewhere); local files must be uploaded via other means beforehand.

Breadth of capabilities is backed by the platform registry: SKILL.md guides users to run `belt app list --category video` / `belt app search` to discover unlisted models; the tables are merely a curated snapshot rather than an absolute ceiling.

tools/video/ai-video-generation/SKILL.md

belt app list --category video

Thus, the '40+ models' actually count around 27 app-ids in the table, with the full catalog defined by the platform runtime.

Complex capabilities (lipsync/avatar/foley/concatenation/upscaling) are conveyed via 'example as minimal viable documentation': Each example provides a complete JSON payload, such as OmniHuman avatars, Hunyuan Foley sound effects, and media-merger concatenation.

tools/video/ai-video-generation/SKILL.md

belt app run infsh/hunyuanvideo-foley --input '{

The skill functions as API documentation through 20+ examples, allowing the model to learn parameter names and composition patterns directly from them.

2Core Capabilities

01Text-to-video: Veo 3.1/3/2, Seedance 2.0, HappyHorse T2V, Grok, P-Video, WAN-T2V

02Image-to-video / Reference-to-video: Wan 2.5, Seedance, HappyHorse I2V/R2V, P-Video

03Avatar / Lipsync: OmniHuman 1.5/1.0, Fabric 1.0, PixVerse Lipsync

04Video editing & post-processing: HappyHorse Edit, Hunyuan Foley audio effects, Topaz upscaling, media-merger concatenation

05Generation with synchronized audio (Seedance generate_audio example)

06Pipeline composition with repository skills (Image → Video → Post, cross-linked via Related Skills)

3External Dependencies

Kind	Dependency / Endpoint
cli	belt（inference.sh CLI）
package	belt-sh/cli（npx skills add 安装）
package	inference-sh/skills 相关技能包（seedance/google-veo/happyhorse/p-video/…）
api	平台视频模型 app（google/veo-3-1-fast、bytedance/seedance-2-0、falai/wan-2-5、xai/grok-imagine-video、bytedance/omnihuman-1-5、alibaba/happyhorse-1-0-、pruna/、infsh/* 等）
network	平台文档（docs/apps/running、docs/api/sdk/streaming 等）
network	安装说明托管

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Graded/per-use authorization, sandbox recommended

For Administrators

Approval workflow, rollback support, full audit

Evidence & Triggers (File Paths & Excerpts)

网络端点: Declared external network access: platform belt CLI → inference.sh platform (video generation/editing apps; media assets hosted on cloud.inference.sh)
SKILL.md: Declared invocation of host CLI interpreter or external runtime environment

Audit Notes: Pure catalog and template-based skill: Contains no scripts, no credential-reading logic, and no direct networking code. All execution converges on running the official `belt` CLI on the host, with predictable external targets (the inference.sh platform and its integrated backend model providers). Key considerations: 1) Each app run consumes user inference.sh account quota (video generation has high per-inference unit costs); 2) Media URLs and prompt contents are transmitted externally to third-party model providers, requiring user discretion regarding likeness and copyright rights; 3) Supply chain: belt and skill packages are distributed via npx without version pinning, and this fork is an unrenamed mirror of inference-sh/skills without a LICENSE file in-tree. Under standard classification guidelines: The `belt` CLI invoked by this skill is distributed via third-party npm/npx (installed via `npx skills add belt-sh/cli`), runs under `belt login` session state, and incurs inference.sh account quota usage/billing. Because it invokes a third-party CLI and touches quota/login state, it is categorized as Amber (Use after evaluation); the skill source code itself contains no token or key extraction routines.

5Second-Pass Verification

[discrepancy] description 宣称 40+ models — 表内可数 app-id 行约 27（含编辑/工具类），'40+ models' 无法在本 SKILL.md 内定位核实——属于平台注册表层面的宣称（belt app list 可查），源码内数字对不上。
[ok] Models 名单（Veo 3.1/3/2、Seedance 2.0、HappyHorse、Wan 2.5、Grok、OmniHuman、Fabric、HunyuanVideo）逐项定位 — 每项均在 Available Models 表有对应 app-id 行；Fabric/HunyuanVideo 以 falai/fabric-1-0、infsh/hunyuanvideo-foley 出现。
[ok] 能力声明（t2v/i2v/ref-to-video/编辑/lipsync/avatar/放大/foley）有示例支撑 — ref-to-video 由 HappyHorse R2V 行与 Seedance 参考示例支撑；foley/lipsync/放大/拼接均有独立示例段。
[ok] 无脚本/无凭证/无直连网络 — 目录仅 SKILL.md；curl/wget/密钥类 token 零命中；执行面只有 belt（allowed-tools 白名单）。
[ok] 媒体输入全远程 URL 判断 — 所有示例的输入键为 image_url/audio_url/video_url/image/video，值均为 https://…，无本地路径/上传指令。
[ok] 执行示例时的本地副作用 — 本文件示例无 shell 重定向（grep '>' 仅命中文档引用行），不写本地文件。
[ok] license/stars/commit 元数据 — pin becc25649700 == 本地 HEAD；101-skills/superpowers stars=0、pushed_at 2026-08-31T22:58:53Z；plugin.json 声明 MIT、repository 指向 inference-sh/skills；树内无 LICENSE。
[unlocatable] installs 安装量 — 无公开安装量数据源，置 null。

6Verdict & Scenarios

Model selection solved in a single table: Mainstream video models are grouped by task with preset 'Best For' judgments, saving the agent from cross-comparing vendors.

Examples serve as documentation: 20+ executable examples with complete JSON cover t2v, i2v, avatars, editing, foley, and concatenation, allowing parameter names to be copied directly.

Whitelisted permissions: allowed-tools: Bash(belt *) narrows host shell authorization down to a single CLI binary.

Comprehensive ecosystem linking: Inter-references with skill packs like seedance, google-veo, happyhorse, ai-avatar-video, ai-image-generation, and twitter-automation to build full content pipelines.

Best for: Ideal for content creators and marketing teams needing to produce AI video directly within agent workflows: an end-to-end chain of copywriting → model selection → generation → (publishing); also suited for users wanting to benchmark visual quality across Veo, Seedance, Wan, and HappyHorse via a single unified CLI.

Not for: Not suitable for scenarios requiring local rendering, offline operation, or self-hosted compute; not suitable for environments with strict requirements against assets leaving the network (must be uploaded to third-party platforms); not suitable for zero-budget automations or scenarios requiring granular budget limits; casual users needing occasional video generation can use official web interfaces directly.

Install Agent-Ready

① Verified Mirror 2026-09-09

Option A · Direct DownloadDownload ai-video-generation.tar.gz

sha256: 7a66e7d4bedac293…

Option B · Copy Guide for Agent

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit becc256497.