---
slug: "claude-media-skills"
source_type: "readme"
source_url: "https://cdn.jsdelivr.net/gh/dennisonbertram/claude-media-skills@main/README.md"
repo: "https://github.com/dennisonbertram/claude-media-skills"
source_file: "README.md"
branch: "main"
---
# Claude Media Skills

A collection of Claude Code skills for AI-powered image and video generation.

## Skills Included

### nano-banana
Generate images using Google's Nano-Banana (Gemini) model. Supports **image-to-image** for character consistency.

**Triggers:** "generate image", "create picture", "make illustration"

**Basic Usage:**
```bash
~/.claude/skills/nano-banana/scripts/generate.sh "your prompt" --aspect 16:9 --size 2K
```

**Image-to-Image (Character Consistency):**
```bash
# First, generate a reference image
~/.claude/skills/nano-banana/scripts/generate.sh "your character description" --output reference.png

# Then generate variations using --reference
~/.claude/skills/nano-banana/scripts/generate.sh "Create this character in a different pose" --reference reference.png --output pose1.png
```

**Options:**
| Flag | Description |
|------|-------------|
| `--output PATH` | Save to specific path |
| `--aspect RATIO` | 1:1, 16:9, 9:16, 4:3, 3:4 |
| `--size SIZE` | 1K, 2K, 4K (Pro model for 2K/4K) |
| `--reference IMAGE` | Reference image for character consistency |
| `--pro` | Use Pro model (higher quality) |
| `--flash` | Use Flash model (faster) |

### realistic-ugc-video
Create realistic long-form AI talking head/UGC videos that don't look AI-generated.

**Triggers:** "realistic AI video", "UGC video", "talking head video", "AI spokesperson"

**Features:**
- Consistent character generation via image-to-video
- Natural pacing with the 55-60 syllable method
- Movement choreography for realistic motion
- Audio generation with Kling 2.6

**Workflow:**
1. Generate base image with Nano Banana
2. Upload to image host (litterbox.catbox.moe works well)
3. Use Kling 2.6 with `image_url` + `generate_audio: True`
4. Same character across all clips!

## Installation

Copy the skill folders to your Claude Code skills directory:

```bash
cp -r nano-banana ~/.claude/skills/
cp -r realistic-ugc-video ~/.claude/skills/
```

## Requirements

- **SEGMIND_API_KEY** - For Kling video generation
- **GEMINI_API_KEY** - For Nano Banana image generation

## Key Learnings

### Pacing for Video
- ~15 words per 5 seconds = natural pacing
- ~30 words per 10 seconds = same ratio
- Too many words = racing speech
- Too few words = awkward pauses

### Character Consistency
- Generate a base image first
- Upload to litterbox.catbox.moe (catbox.moe returns empty responses)
- Use `image_url` parameter in Kling 2.6
- Same image = same character across clips

### Pronunciation
- Avoid slang: "Lemmie" confuses the model
- Split compound words: "Git Hub" instead of "GitHub"
- Use phonetic spelling for tricky words

## License

MIT