# Create AI Videos with Inference.sh

Video teams need fast access to many AI video models without memorizing each provider API. This skill guides Claude, Codex, and Claude Code through inference.sh belt commands for generation, editing, upscaling, and audio workflows.

## Install

```bash
npx skillstore add inference-skills/ai-video-generation
```

## Metadata

- Status: approved
- Slug: inference-skills-ai-video-generation
- Skillstore revision: r1
- Version status: missing
- Tree hash: c761a31f25d863b6eaec7d754ee723e340c7344114abf26774288c4c8ce2e663
- Author: inference-skills
- GitHub username: inference-skills
- License: MIT
- Repository: https://github.com/inference-skills/skills/tree/main/tools/video/ai-video-generation/
- Ref: a25199bc7d6b82598536822d1738eb5d5f54025b
- 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
- Quality score: 69
- Public page: https://skillstore.pages.dev/skills/inference-skills-ai-video-generation
- Manifest: https://skillstore.pages.dev/api/skills/inference-skills-ai-video-generation/manifest

## Capabilities

- Lists inference.sh video model choices for text-to-video, image-to-video, avatar, lipsync, editing, and utility tasks.
- Shows belt CLI commands for running Google Veo, Seedance, HappyHorse, Wan, Grok, OmniHuman, Fabric, PixVerse, and utility apps.
- Supports prompt-based video generation with options such as duration, resolution, audio generation, and reference media.
- Guides image, audio, and video URL inputs for animation, lipsync, foley sound, upscaling, and clip merging workflows.
- Points users to inference.sh documentation for running apps, streaming results, and content pipeline examples.

## Use Cases

- Create campaign clips: Generate short product, social, or brand videos from a written concept and selected model settings.
- Prototype video concepts: Compare text-to-video and image-to-video models quickly before committing to a production workflow.
- Build avatar media: Animate portraits, sync speech audio, add sound effects, and upscale final clips for presentation use.

## Prompt Templates

### Generate a simple video

```
Use the ai-video-generation skill to create a five second video of a calm ocean sunrise. Choose a fast model and explain any account or input requirements before running it.
```

### Animate a product image

```
Use this product image URL to create an image-to-video clip with subtle camera movement and clean studio lighting. Recommend a suitable model and duration.
```

### Create a talking avatar

```
Create a talking head video from a portrait URL and an audio URL. Select an avatar or lipsync model and include any privacy checks before submission.
```

### Plan a full video pipeline

```
Design a workflow that generates three short clips, adds foley sound, merges them with transitions, and upscales the result. Ask for confirmation before running each paid or external command.
```

## Limitations

- Requires the inference.sh belt CLI and an authenticated account before commands can run.
- Generated results depend on third-party model availability, pricing, rate limits, and provider policies.
- The skill does not verify copyright, likeness rights, consent, or commercial usage rights for submitted media.
- Examples use placeholder URLs and prompts that users must replace with valid accessible media.

## Best Practices

- Confirm account, cost, and privacy expectations before running belt commands.
- Use media only when you have rights and consent for the people, voices, brands, and source files involved.
- Start with short durations and lower-cost models before producing final high-resolution clips.

## Anti Patterns

- Do not submit private, regulated, or confidential media without explicit approval.
- Do not run generated commands that install other skills unless the user asked for that action.
- Do not assume model outputs are legally cleared for commercial use without checking provider terms.

## Security Audit

- Audited at: 2026-07-05T17:05:12.113\+00:00
- Summary: The static external-command hits are Markdown fences, inline model identifiers, or example commands rather than Ruby backtick execution. The hardcoded URL hits are documentation links, a rendered image, or placeholder media URLs, but the skill does grant broad belt CLI execution that can send user prompts and media references to external services. No prompt injection or hidden exfiltration intent was found in SKILL.md.

## Stats

- Views: 61
- Downloads: 6
- Favorites: 0
- Popularity score: 0