# Gemini Skills for Agents
Claude Code and Codex plugin for Gemini-powered image generation, video generation, editing, and visual analysis. Powered by **Gemini 3.1 Pro** (`gemini-3.1-pro-preview`), **Nano Banana Pro** (`gemini-3-pro-image`), **Veo 3.1** (`veo-3.1-generate-preview`), and **Gemini 3 Flash**. 169 art styles, text-to-video, image-to-video, pixel avatars, presentation decks, and more.
Every image and video on this page was generated using gemskills.
## Installation
### Claude Code
```bash
/plugin marketplace add b-open-io/claude-plugins
/plugin install gemskills@b-open-io
```
The Claude plugin includes Lisa, the `content` agent, natively.
### Codex
Install Gemskills from the b-open-io Codex marketplace:
```bash
codex plugin marketplace add b-open-io/gemskills --ref master
codex plugin add gemskills@b-open-io
```
The plugin exposes its skills and the explicit
`$gemskills:codex-agent-setup` installer skill. Codex plugin installation alone
does **not** install custom agents.
When you explicitly want Lisa available to Codex, invoke the setup skill or run:
```bash
bash skills/codex-agent-setup/scripts/setup.sh # current project
bash skills/codex-agent-setup/scripts/setup.sh --user # user scope, explicit only
```
The installer copies `gemskills-content.toml` as a regular file and never edits
global Codex configuration. Start a new Codex session after installation, then
invoke Lisa with the runtime agent name `gemskills_content`.
Individual skills (for other agentic frameworks)
```bash
bunx skills add b-open-io/gemskills --skill generate-image
bunx skills add b-open-io/gemskills --skill generate-video
bunx skills add b-open-io/gemskills --skill browsing-styles
bunx skills add b-open-io/gemskills --skill avatar-portrait
bunx skills add b-open-io/gemskills --skill team-group-photo
bunx skills add b-open-io/gemskills --skill generate-icon
bunx skills add b-open-io/gemskills --skill edit-image
bunx skills add b-open-io/gemskills --skill upscale-image
bunx skills add b-open-io/gemskills --skill segment-image
bunx skills add b-open-io/gemskills --skill optimize-images
bunx skills add b-open-io/gemskills --skill generate-svg
bunx skills add b-open-io/gemskills --skill section-dividers
bunx skills add b-open-io/gemskills --skill deck-creator
bunx skills add b-open-io/gemskills --skill ask-gemini
bunx skills add b-open-io/gemskills --skill setup
```
**Requirements**: `GEMINI_API_KEY` ([get one](https://aistudio.google.com/apikey)) is the baseline and powers every skill.
Optional keys unlock additional providers for image/video/edit:
| Key | Unlocks | Get one |
|-----|---------|---------|
| `GEMINI_API_KEY` | Gemini Nano Banana Pro images, Veo 3.1 video, all 169 styles (default) | [aistudio.google.com](https://aistudio.google.com/apikey) |
| `OPENAI_API_KEY` | OpenAI **`gpt-image-2`** image generation + masked editing | [platform.openai.com](https://platform.openai.com/api-keys) |
| `XAI_API_KEY` | xAI **Grok Imagine** image + **`grok-imagine-video-1.5`** video | [console.x.ai](https://console.x.ai) |
| `REPLICATE_API_TOKEN` | Icon background removal; Veo reference-image / last-frame video | [replicate.com](https://replicate.com/account/api-tokens) |
### Providers & auto-pick
`generate-image`, `generate-video`, and `edit-image` accept `--provider gemini|openai|xai`.
Omit it and gemskills **auto-picks the best provider whose key is present and that
supports the request** — e.g. plain image → `gpt-image-2`, video → `grok-imagine-video-1.5`,
but anything needing **style tiles, reference images, transparency, or negative
prompts routes to Gemini** (the only provider that supports them).
Set your own defaults interactively with **`/gemskills:setup`** (or the `setup`
skill), or pin them via env (`GEMSKILLS_IMAGE_PROVIDER`, `GEMSKILLS_VIDEO_PROVIDER`,
`GEMSKILLS_EDIT_PROVIDER`) or a `.gemskills.json` (project) / `~/.config/gemskills/config.json`
(global) file. Per-provider prompt templates live in `providers/prompts/` and are
tuned independently. Keys are read from one canonical env var each — gemskills
never falls back to alternate names; a missing key fails loudly.
---
## Video Generation
Generate videos from text or animate existing images with Veo 3.1. Native audio, 720p/1080p/4K, 4-8 second clips, and all 169 art styles.
### Style + Prompt → Image + Prompt → Video
| Style |
→ |
Generated Image |
→ |
Generated Video |
--style kusm |
+prompt |
Nano Banana Pro |
+prompt |
Veo 3.1 · 8s |
--style impr |
+prompt |
auto-generated |
+prompt |
Veo 3.1 · 8s |
The two-step pipeline gives full creative control: generate a styled image first, then animate it. Or use `--auto-image` to do both in one command.
### Text-to-Video
Prompt alone → complete scene with audio. No input image needed.
"Ocean waves crashing on dark volcanic rocks at golden hour..." · 8s · ~67s gen
> [Text-to-video demo](skills/generate-video/examples/text-to-video/prompt.md) | [Image-to-video demo](skills/generate-video/examples/image-to-video/prompt.md) | [Auto-image pipeline demo](skills/generate-video/examples/auto-image-pipeline/prompt.md)
---
## Image Generation
### 169 Art Styles
Every image and video generation supports `--style` to apply any of 169 curated art styles. Each style includes an AI-generated tile reference image sent to Gemini alongside the prompt for dramatically better style adherence.
Kusama kusm |
H.R. Giger gigr |
Spider-Verse spdr |
Ukiyo-e ukiy |
Murakami mrkm |
Stained Glass stgl |
Wes Anderson wesa |
Studio Ghibli ghbl |
Impressionism impr |
Cyberpunk cybr |
Art Deco deco |
Vaporwave vapr |
Made of Sand sand |
Claymation clay |
Film Noir fnoi |
Steampunk stpk |
Pixel Art pixl |
Solarpunk solr |
> **[View all 169 styles with categories and short codes](STYLES.md)**
### Styled Avatars & Team Photos
Transform headshots into styled portraits, then composite into group scenes.
Maya
Creative Director |
Kai
Lead Engineer |
Yuki
Art Director |
Carlos
Audio Director |
 |
 |
 |
 |
 |
 |
 |
 |
> [Avatar portrait demo](skills/avatar-portrait/examples/team-portraits/prompt.md) | [Team group photo demo](skills/team-group-photo/examples/group-composite/prompt.md)
### Social Share & App Icons
Generate OG images cropped to platform specs, or production-ready icon sets for iOS, Android, web, and desktop from a single prompt.
| Input |
OG 1200x630 |
Master Icon |
No BG |
App Store |
 |
 |
 |
 |
 |
> [Social share demo](skills/generate-image/examples/social-share/prompt.md) | [App icon demo](skills/generate-icon/examples/gemskills-app-icon/prompt.md)
---
## Presentations
Generate complete pitch decks with consistent visual style, then restyle with any of the 169 art styles.
| 01 Title |
02 Problem |
05 Portfolio |
09 Traction |
 |
 |
 |
 |
Same deck, restyled with --style pixl |
 |
 |
 |
 |
> [Pitch deck demo](skills/deck-creator/examples/opl-pitch-deck/prompt.md) | [Pixel art variant](skills/deck-creator/examples/opl-pitch-deck-pixelart/prompt.md)
---
## Design & Prompt Refinement
Use ask-gemini for prompt refinement, design critique, and full page redesigns from HTML + inspiration screenshot.
**Rough concept** → **Gemini refinement** → **Generated image** (the hero image above was made this way)
**HTML + inspiration** → **Redesigned code** ([input](skills/ask-gemini/examples/design-feedback/inputs/current-page.html) → [output](skills/ask-gemini/examples/design-feedback/outputs/updated-page.html))
> [Prompt refinement demo](skills/ask-gemini/examples/prompt-refinement/prompt.md) | [Design redesign demo](skills/ask-gemini/examples/design-feedback/prompt.md)
---
## All Skills
| Skill | Description |
|-------|-------------|
| **generate-image** | AI image generation with [169 art styles](STYLES.md), multi-reference (up to 14 images), img2img |
| **generate-video** | Text-to-video, image-to-video with Veo 3.1, native audio, auto-image pipeline |
| **browsing-styles** | Browse, search, and preview all 169 art styles |
| **avatar-portrait** | Likeness-preserving avatar portraits in any requested style |
| **pixel-avatar** | Compatibility alias for pixel-style avatar requests |
| **team-group-photo** | Individual styled portraits composited into group scenes |
| **generate-icon** | Platform icons (favicon, iOS, Android, PWA, desktop) with auto sizing |
| **edit-image** | Inpainting and outpainting with masks |
| **upscale-image** | 2x/4x upscaling via Vertex AI |
| **segment-image** | Object identification and extraction |
| **optimize-images** | Batch compress PNGs/JPEGs for web using sharp |
| **generate-svg** | Vector graphics, logos, and icons |
| **section-dividers** | Transparent decorative dividers for web sections |
| **deck-creator** | Complete presentation decks with consistent visual style |
| **ask-gemini** | Text + image queries for design critique, prompt refinement, spatial analysis |
## Quick Examples
```bash
# Generate an image with art style
"Generate a mountain landscape in watercolor style"
# Generate a video from text
"Generate a video of ocean waves crashing on rocks"
# Animate an image into video
"Turn this image into a video with gentle motion"
# Create styled avatar from photo
"Create a stylized avatar portrait from my headshot"
# Ask Gemini to refine a prompt
"Ask Gemini to write a better prompt for: a futuristic city at sunset"
# Create presentation slides
"Create a pitch deck for my startup"
# Generate app icons for all platforms
"Generate a favicon for my website with a lightning bolt"
```
## Why Gemini
- **Nano Banana Pro**: Google's latest image generation model with thinking capabilities
- **Veo 3.1**: Text-to-video and image-to-video with native audio generation
- **[169 art styles](STYLES.md)**: Curated style library with AI-generated tile references for visual adherence
- **Multi-image input**: Up to 14 reference images for character consistency and scene composition
- **Spatial reasoning**: Superior visual understanding for design feedback
- **Dynamic docs**: Fetch latest API docs via [llms.txt](https://ai.google.dev/gemini-api/docs/llms.txt)
## License
MIT