# Create AI Avatar Videos with RunComfy

Producing convincing avatar video requires selecting compatible models, inputs, and prompts. This skill routes each request and prepares RunComfy CLI commands for talking heads, lip sync, and cinematic presenters.

## Install

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

## Metadata

- Status: approved
- Slug: agentspace-so-ai-avatar-video
- Skillstore revision: r2
- Version status: missing
- Tree hash: 18f60ddd2d3f0571b9c1c4ddd0267499a60c9cad6a2a0dc7c85d6b09decb8007
- Author: agentspace-so
- GitHub username: agentspace-so
- License: MIT
- Repository: https://github.com/agentspace-so/runcomfy-agent-skills/tree/main/ai-avatar-video/
- Ref: 181fdefcafd96b041926e61c4b2e306ca7e7820e
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/agentspace-so-ai-avatar-video
- Manifest: https://skillstore.pages.dev/api/skills/agentspace-so-ai-avatar-video/manifest

## Capabilities

- Routes requests across OmniHuman, Wan 2-7, Wan 2-2 Animate, HappyHorse 1.0, and Seedance v2 Pro.
- Builds RunComfy CLI invocations for portrait, audio, prompt, duration, resolution, and aspect ratio inputs.
- Supports audio-driven portraits, script-driven speech, stylized character animation, and multimodal cinematic shots.
- Provides model-specific prompting guidance for speech, emotion, camera movement, lighting, and reference assets.
- Explains CLI exit codes, output downloads, authentication, consent requirements, and common troubleshooting paths.

## Use Cases

- Produce Product Presenter Clips: Create short portrait-led product videos from approved voiceovers and brand imagery.
- Localize Presenter Content: Reuse one approved portrait with separate audio tracks for multiple language versions.
- Animate Illustrated Characters: Pair a mascot or illustrated character with dialogue and synchronized full-body movement.

## Prompt Templates

### Create a Basic Talking Head

```
Create a talking-head video from portrait [image URL] and voiceover [audio URL]. Choose the simplest suitable model and explain required inputs.
```

### Generate Speech from a Script

```
Turn this script into a six-second vertical presenter video: [script]. Use a friendly tone, neutral background, and direct eye contact.
```

### Animate a Stylized Character

```
Animate [character image URL] with [audio URL]. Preserve the illustrated style, synchronize full-body motion, and select the best character model.
```

### Plan a Cinematic Monologue

```
Plan a cinematic monologue using [subject image], [voiceover], and [scene references]. Specify model, duration, aspect ratio, camera movement, lighting, and fallback route.
```

## Limitations

- Requires the RunComfy CLI, a RunComfy account, network access, and available API credits.
- Sends prompts and referenced media to an external generation service, using URLs that the service can access.
- Does not verify identity, likeness rights, voice rights, or consent for supplied media.
- Output quality, timing, lip synchronization, model availability, and input schemas can change across service versions.

## Best Practices

- Obtain documented consent and usage rights for every face, voice, image, and audio source.
- Use clean speech audio, clear portraits, short scripts, and model-appropriate framing.
- Review the current model schema, expected cost, privacy terms, and generated video before publication.

## Anti Patterns

- Do not imitate real people or clone voices without explicit, verifiable consent.
- Do not place untrusted user text directly into shell command strings.
- Do not use a complex cinematic model for a basic portrait-and-audio request.

## Security Audit

- Audited at: 2026-07-23T07:41:11.485\+00:00
- Summary: Most alerts are false positives caused by Markdown backticks, documentation links, placeholder URLs, model guidance, and standard token-storage documentation. Three findings remain confirmed because the skill recommends unpinned global npm or npx execution. A separate high-severity issue exists where user-derived content can break single-quoted shell arguments before the CLI receives them.

## Stats

- Views: 53
- Downloads: 7
- Favorites: 0
- Popularity score: 0