Introduction

When developing video generation applications, a common pain point is that results vary significantly across different models for the same task description, and no team wants to integrate every available video model just to trial and error them one by one. The “Video Model Guide” Skill works by having the Agent first classify the task (e.g., image-to-video, start-end frames, cinematic camera movement) before generation, and then dispatches the corresponding model. The following section outlines the positioning, capabilities, and usage of this Skill.

What is this

Video Model Guide (@infimind/video-creator) is a Skill categorized under ai-agent, published by JiRui Technology - AI Experts and tagged as “enterprise” origin. The current version is 1.0.0. It addresses the issue of how an Agent selects and dispatches video models such as Happyhorse, Seedance, and Kling when receiving a video generation task, rather than requiring the user to manually determine the matching relationship between models and tasks.

Core Functions

According to the official description, it has three main capabilities:

  • Model Scheduling: The Agent intelligently schedules video models such as Happyhorse, Seedance, and Kling.
  • Task Matching: Automatically matches the optimal model based on three task categories: “Image-to-Video / Start-End Frames / Cinematic Camera Movement.”
  • Video Production Capabilities: Covers six types of finished products: product explanations, training screen recordings, promotional shorts, model runway shows, multi-shot storyboards, and meeting summary videos. Supports multiple duration options from 5-15 seconds, with aspect ratios of 9:16, 1:1, and 16:9.

The page tag requires_api_key=true indicates that the corresponding model API key must be configured for actual invocation.

Installation and Activation

The official page does not provide specific installation commands. This Skill is published in the standard SKILL.md format with the coordinate @infimind/video-creator and is obtained from the SkillHub page. Before activation, you need to prepare the API keys for the video models being called. The official documentation does not provide separate instructions for installation in Cursor, Codex CLI, or Claude Code, so no speculation is made here.

Typical Usage

The following usage scenarios are derived from the task types described in the official introduction. No executable command examples were found in the materials:

  • Image-to-Video: Generates dynamic video starting from a single image, suitable for product explanations, promotional shorts, and model runway shows.
  • Start-End Frames: Constrains the starting and ending frames of a video using two images (the first and last frame), suitable for segments requiring clear shot beginnings and endings.
  • Cinematic Camera Movement: Generates shot segments with camera movement, suitable for multi-shot storyboards and meeting summary videos.
  • Duration and Aspect Ratio: Multiple duration options from 5-15 seconds; supports 9:16, 1:1, and 16:9 aspect ratios, selectable based on the distribution channel.

Use Cases and Notes

This is suitable for teams that have already integrated Agent workflows and need to perform task matching across multiple video models. Regarding limitations: the official introduction does not specify details such as pricing, concurrency limits, or review rules for each model; the SkillHub page shows a rating of 100000 and 1246 downloads (data at the time of scraping), with 0 installs. It is recommended to validate whether the model matching meets expectations using small-scale tasks before full integration.

Summary

The value of the Video Model Guide lies in handing over the “Task Type → Model” matching logic to the Agent, rather than making the user try models one by one.
Official Address: https://api.skillhub.cn/infimind/video-creator (SkillHub Page: https://www.skillhub.cn/skills/infimind/video-creator).