# Create Avatar and Lip-Sync Videos with RunComfy

Choosing an avatar model depends on available images, audio, and scene requirements. This skill guides model selection and RunComfy CLI inputs for five video-generation routes.

## Install

```bash
npx skillstore add prime-skills/ai-avatar-video
```

## Metadata

- Status: approved
- Slug: prime-skills-ai-avatar-video
- Skillstore revision: r1
- Version status: missing
- Tree hash: 18f60ddd2d3f0571b9c1c4ddd0267499a60c9cad6a2a0dc7c85d6b09decb8007
- Author: prime-skills
- GitHub username: prime-skills
- License: MIT
- Repository: https://github.com/prime-skills/runcomfy-agent-skills/tree/fca19ae084c2ae4667f5c8affbb99da5103d5832/ai-avatar-video
- Ref: 2db34eaa94f4b8d5bbd0bb0d5c6d624841357495
- 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: 69
- Public page: https://skillstore.pages.dev/skills/prime-skills-ai-avatar-video
- Manifest: https://skillstore.pages.dev/api/skills/prime-skills-ai-avatar-video/manifest

## Capabilities

- Select among five documented routes based on recorded audio, script availability, subject style, and scene complexity.
- Prepare OmniHuman inputs pairing a portrait image with a recorded voiceover.
- Prepare Wan 2-7 scene prompts with an audio track, or Wan 2-2 Animate inputs for stylized characters.
- Write HappyHorse text-to-video and image-to-video prompts with quoted dialogue and separate audio-tone descriptions.
- Prepare Seedance cinematic prompts with reference images, reference audio, duration, and aspect ratio.
- Document RunComfy installation, authentication, output-directory selection, and common CLI exit codes.

## Use Cases

- Localized Presenter Clips: Prepare OmniHuman jobs using one authorized presenter portrait and separate voiceovers for each language.
- Script-Based Training Videos: Prepare short HappyHorse presenter clips from written dialogue when recorded audio is unavailable.
- Animated Character Dialogue: Prepare Wan 2-2 Animate inputs pairing an illustrated character with an authorized voice recording.

## Prompt Templates

### Choose an Avatar Model

```
I have an authorized portrait and recorded voiceover for a presenter clip. Recommend the simplest documented route and list the required inputs.
```

### Prepare a Talking Portrait

```
Use [portrait URL] and [voiceover URL] for an OmniHuman presenter clip. Prepare the RunComfy invocation with safe argument handling. Do not submit it yet.
```

### Create a Script-Based Clip

```
Draft a six-second vertical HappyHorse clip saying [short script] with a friendly tone. Show the prompt and generation settings before submission.
```

### Plan a Cinematic Monologue

```
Plan a ten-second Seedance monologue using [subject URL] and [audio URL]. Include lens, lighting, and aspect ratio. Verify input fields and await submission approval.
```

## Limitations

- Requires a separately installed RunComfy CLI, authentication, network access, and access to the selected hosted model.
- Provides instructions and examples, not a bundled video engine or independently verified CLI security controls.
- Asset-based routes require reachable image or audio URLs; supplied examples contain placeholders.
- Generated speech, lip synchronization, and identity consistency depend on the model; current schemas and service charges require confirmation.

## Best Practices

- Obtain permission for every likeness and voice, and disclose synthetic media where appropriate.
- Use clean voice recordings and well-framed portraits; confirm the current input schema before submitting a generation.
- Pass inputs through structured arguments or robust shell escaping, protect credentials, and review external processing and costs before submission.

## Anti Patterns

- Using a real person's likeness or voice without permission, or creating harmful synthetic media.
- Choosing HappyHorse when exact synchronization to an existing recording is required.
- Pasting unescaped prompts into shell commands or assuming JSON encoding alone prevents command injection.

## Security Audit

- Audited at: 2026-11-14T14:43:26.469\+00:00
- Summary: All 93 static matches are false positives: Markdown formatting, relevant documentation links, placeholder asset URLs, ordinary credential-storage documentation, and model-selection advice. One semantic concern remains: the shell-safety assurance overlooks quoting when inserting user content into single-quoted command examples. No evidence found of malicious exfiltration, system reconnaissance, or prompt injection targeting the auditor.

## Stats

- Views: 0
- Downloads: 2
- Favorites: 0
- Popularity score: 0
