# Generate AI Videos with RunComfy

Choosing a video model and its required inputs can slow production. This skill routes each request and prepares a focused RunComfy generation command.

## Install

```bash
npx skillstore add agentspace-so/ai-video-generation
```

## Metadata

- Status: approved
- Slug: agentspace-so-ai-video-generation
- Skillstore revision: r2
- Version status: missing
- Tree hash: 98fbb84b4eb09f5a24e429bfa7efb02d6d954259be9e45a4899423f83dc30453
- Author: agentspace-so
- GitHub username: agentspace-so
- License: MIT
- Repository: https://github.com/agentspace-so/runcomfy-agent-skills/tree/main/ai-video-generation/
- Ref: 181fdefcafd96b041926e61c4b2e306ca7e7820e
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 78
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/agentspace-so-ai-video-generation
- Manifest: https://skillstore.pages.dev/api/skills/agentspace-so-ai-video-generation/manifest

## Capabilities

- Classifies requests as text-to-video, image-to-video, or video extension.
- Selects documented RunComfy models based on quality, speed, audio, motion, duration, and cost needs.
- Builds model-specific prompts with camera, motion, audio, aspect ratio, duration, and resolution guidance.
- Invokes RunComfy endpoints through the runcomfy CLI and saves generated media to a chosen output directory.
- Provides patterns for lip-sync, product motion, vertical clips, cinematic ads, and multi-shot character scenes.

## Use Cases

- Create vertical social clips: Turn a short concept or still image into a vertical video with planned motion, audio, duration, and framing.
- Develop cinematic campaign shots: Choose a premium model and structure camera, lens, reference, identity, and audio direction for a polished advertising shot.
- Automate video generation: Route repeatable requests to suitable RunComfy endpoints and save generated clips to a defined output directory.

## Prompt Templates

### Create a text-to-video clip

```
Generate a {duration}-second {aspect_ratio} video of {subject} {action} in {setting}. Include {camera_motion}, {visual_style}, and {audio_details}. Choose a suitable RunComfy model.
```

### Animate a still image

```
Animate the user-provided image at {image_url}. Show {main_motion}, keep {locked_elements} unchanged, use {camera_behavior}, and deliver {duration} seconds in {aspect_ratio}.
```

### Generate a synchronized dialogue scene

```
Create a {duration}-second scene featuring {speaker_description} in {setting}. Use the approved audio at {audio_url} for lip-sync and keep the camera {camera_direction}.
```

### Direct a multi-reference campaign shot

```
Generate a {duration}-second {aspect_ratio} campaign video using {subject_references}, {scene_references}, and {audio_references}. Preserve {identity_constraints}, follow {shot_sequence}, and target {delivery_quality}. Explain the model choice.
```

## Limitations

- Requires the RunComfy CLI, valid authentication, network access, and access to the selected hosted model.
- Model availability, pricing, latency, schemas, and output quality can change outside the skill.
- Reference media must use accessible URLs because local upload handling is not documented.
- Generated videos can contain artifacts and require human review for rights, consent, accuracy, and brand compliance.

## Best Practices

- Provide one main action, exact duration, aspect ratio, and intended camera behavior.
- Use only reference URLs that you trust and explicitly approve for the current generation.
- Review model costs and generate lower-cost drafts before producing final high-resolution clips.

## Anti Patterns

- Do not include API tokens, private credentials, or confidential data in prompts or media URLs.
- Do not request real-person impersonation or face use without appropriate rights and consent.
- Do not assume every model supports the same fields, durations, resolutions, or audio behavior.

## Security Audit

- Audited at: 2026-07-23T08:30:22.818\+00:00
- Summary: All 154 static findings are false positives caused by Markdown formatting, CLI examples, reference links, standard configuration paths, and model-selection prose. The skill intentionally invokes RunComfy and sends generation inputs to its API. No malware, prompt injection, credential exfiltration, or arbitrary command execution was found.

## Stats

- Views: 61
- Downloads: 9
- Favorites: 0
- Popularity score: 0