# Animate Still Images with RunComfy Models

Choosing the wrong image-to-video model wastes time and generation credits. This skill routes each request to a suitable RunComfy model and supplies model-specific inputs.

## Install

```bash
npx skillstore add agentspace-so/image-to-video
```

## Metadata

- Status: approved
- Slug: agentspace-so-image-to-video
- Skillstore revision: r2
- Version status: missing
- Tree hash: 9c50b2b32907fb02cb6ae3b02f8a7328ca6eaf4054b376100afa5591b5492007
- Author: agentspace-so
- GitHub username: agentspace-so
- License: MIT
- Repository: https://github.com/agentspace-so/runcomfy-agent-skills/tree/main/image-to-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-image-to-video
- Manifest: https://skillstore.pages.dev/api/skills/agentspace-so-image-to-video/manifest

## Capabilities

- Routes general animation requests to HappyHorse 1.0 I2V.
- Routes custom voiceover and lip-sync requests to Wan 2.7.
- Routes combined image, video, and audio references to Seedance 2.0 Pro.
- Provides model-specific prompt, duration, resolution, and aspect-ratio guidance.
- Provides RunComfy CLI invocation patterns for each supported route.
- Explains exit codes, authentication, output downloads, and model limits.

## Use Cases

- Animate a Product Still: Create a short product reveal while preserving packaging geometry, logo placement, and the original image aspect.
- Produce Localized Talking Heads: Generate repeated talking-head clips by changing the voiceover audio while keeping visual settings and seed consistent.
- Combine Multimodal References: Create one clip using a subject image, camera movement from video, and voice tone from audio.

## Prompt Templates

### Animate a Portrait

```
Animate [image URL] as a five-second portrait clip. Keep the face stable. Add subtle breathing, one blink, and a slow camera push-in.
```

### Create a Product Reveal

```
Create a product reveal from [image URL]. Preserve packaging and logo geometry. Use one slow orbit, controlled highlights, 1080P, and eight seconds.
```

### Lip-Sync a Voiceover

```
Create a vertical talking-head clip using [audio URL]. Match the clip duration to the audio. Use warm studio lighting and avoid subtitles or flicker.
```

### Direct a Multimodal Scene

```
Combine [subject image], [reference video], and [reference audio]. Keep the subject identity, follow the reference camera movement, and match the voice tone. Produce an eight-second 9:16 clip with synchronized audio.
```

## Limitations

- Requires the RunComfy CLI, an account, network access, and valid authentication.
- Sends prompts and referenced media to an external model service.
- Selects one model per call and does not blend routes or stitch clips.
- Each route has model-specific duration, resolution, format, and aspect-ratio limits.

## Best Practices

- Use media URLs you are authorized to share with RunComfy.
- Describe one primary motion beat and state the visual details that must remain stable.
- Match clip duration to reference audio and reuse seeds when comparing variants.

## Anti Patterns

- Do not place credentials, private tokens, or unnecessary personal data in media URLs or prompts.
- Do not request multiple model routes in one call because the skill selects only one route.
- Do not combine reference media with conflicting subjects, camera styles, or visual aesthetics.

## Security Audit

- Audited at: 2026-07-23T08:57:55.627\+00:00
- Summary: The static alerts for backtick execution, path traversal, hidden-file access, hardcoded URLs, and reconnaissance are false positives caused by Markdown and documentation examples. The skill still presents shell interpolation, unpinned package installation, and third-party media disclosure risks. No prompt injection attempt was found.

## Stats

- Views: 39
- Downloads: 5
- Favorites: 1
- Popularity score: 0