# Agent Brush

> Image editing toolkit for AI agents. Use when processing, editing, or validating images. Background removal, compositing, text overlays, resizing, format conversion, and spec validation against presets (social media, icons, thumbnails).

- **Type:** Skill
- **Install:** `agentstack add skill-ultrathink-art-agentbrush-agent-brush`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ultrathink-art](https://agentstack.voostack.com/s/ultrathink-art)
- **Installs:** 0
- **Category:** [Content & Media](https://agentstack.voostack.com/c/content-and-media)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [ultrathink-art](https://github.com/ultrathink-art)
- **Source:** https://github.com/ultrathink-art/agentbrush/tree/main/skill/agent-brush

## Install

```sh
agentstack add skill-ultrathink-art-agentbrush-agent-brush
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# AgentBrush — Image Editing for AI Agents

Image processing toolkit with background removal, green screen processing, border cleanup, text rendering, compositing, resizing, validation, and format conversion.

## Quick Reference

All commands follow: `agentbrush    [options]`

### Background Removal (edge-based flood fill — NEVER threshold-based)

```bash
# Remove black background
agentbrush remove-bg input.png output.png --color black --threshold 25

# Remove white background with edge smoothing
agentbrush remove-bg input.png output.png --color white --smooth

# Remove custom color background
agentbrush remove-bg input.png output.png --color "24,242,41" --threshold 40
```

### Green Screen Removal (multi-pass pipeline)

```bash
# Standard green screen removal
agentbrush greenscreen input.png output.png

# With upscale + halo cleanup
agentbrush greenscreen input.png output.png --upscale 3 --halo-passes 20
```

### Border Cleanup (edge artifact removal)

```bash
# Remove white border artifacts from AI-generated images
agentbrush border-cleanup input.png output.png --passes 15 --threshold 185

# Full cleanup: white border + green halo + alpha smoothing
agentbrush border-cleanup input.png output.png --passes 15 --green-halo-passes 20 --alpha-smooth
```

### Text Rendering (Pillow-based, accurate)

```bash
# Add text to existing image
agentbrush text input.png output.png "HELLO WORLD" --font mono --bold --size 72 --color "255,255,255,255"

# Render text on new canvas (use new:WxH as input)
agentbrush text new:1200x630 output.png "Title Text" --font mono --bold --size 80 --center
```

### Compositing

```bash
# Overlay image onto base
agentbrush composite base.png overlay.png output.png --position 100,200

# Center artwork on canvas
agentbrush composite paste-centered output.png --overlay artwork.png --canvas 1200x630 --fit
```

### Resize

```bash
# Resize to exact dimensions
agentbrush resize input.png output.png --width 1200 --height 630

# Scale by factor
agentbrush resize input.png output.png --scale 3.0

# Fit within bounds preserving aspect ratio
agentbrush resize input.png output.png --width 1080 --height 1080 --fit

# Fit and pad to exact dimensions
agentbrush resize input.png output.png --width 1200 --height 630 --pad
```

### Validation (preset-based)

```bash
# Validate against built-in presets
agentbrush validate check image.png --preset social-og
agentbrush validate check image.png --preset favicon
agentbrush validate check image.png --preset icon-ios

# Validate with custom spec
agentbrush validate check image.png --width 800 --height 600 --transparent

# Compare source vs processed (verify bg removal didn't damage artwork)
agentbrush validate compare source.png processed.png --max-loss 10
```

### Format Conversion

```bash
agentbrush convert input.png output.jpg
agentbrush convert input.png output.webp --quality 90
```

### AI Image Generation (requires OpenAI key)

```bash
agentbrush generate --provider openai --prompt "cute cat coding" --output cat.png
```

## Python API

```python
from agentbrush import (
    remove_background,
    remove_greenscreen,
    cleanup_border,
    add_text, render_text,
    composite, paste_centered,
    resize_image,
    validate_design, compare_images,
    convert_image,
)

# All functions return a Result object
result = remove_background("in.png", "out.png", color="black")
print(result.success)     # True/False
print(result.summary())   # Human-readable output
print(result.width)       # Output dimensions
print(result.transparent_pct)  # Transparency percentage
```

## Available Presets

| Preset | Width | Height | Use Case |
|--------|-------|--------|----------|
| `social-og` | 1200 | 630 | Open Graph / link previews |
| `social-square` | 1080 | 1080 | Instagram, social posts |
| `social-story` | 1080 | 1920 | Stories, reels, vertical |
| `favicon` | 32 | 32 | Browser favicon |
| `icon-ios` | 1024 | 1024 | iOS app icon |
| `icon-android` | 512 | 512 | Android app icon |
| `thumbnail` | 400 | 400 | Thumbnails, previews |
| `banner` | 1920 | 480 | Website/profile banners |
| `avatar` | 256 | 256 | Profile avatars |

## Best Practices

1. **NEVER use threshold-based background removal** — destroys internal outlines/details. Always use edge-based flood fill (`remove-bg`).
2. **Use Pillow for text, not AI generation** — AI mangles long text. Use `agentbrush text` for accurate rendering.
3. **Green screen technique**: prompt AI for "#00FF00 background", then use `agentbrush greenscreen` for clean removal.
4. **Validate before delivery** — use `agentbrush validate check` with the appropriate preset to catch dimension/format issues early.
5. **Compare after bg removal** — `agentbrush validate compare` catches artwork damage from processing.

## Additional Resources

- For common issues and fixes, see [references/troubleshooting.md](references/troubleshooting.md)

## Standalone Scripts

The `scripts/` directory contains standalone wrappers for each command. These work without `pip install` — they auto-detect the `src/` directory:

```bash
python scripts/remove_bg.py input.png output.png --color black
python scripts/validate.py check image.png --preset social-og
```

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ultrathink-art](https://github.com/ultrathink-art)
- **Source:** [ultrathink-art/agentbrush](https://github.com/ultrathink-art/agentbrush)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-ultrathink-art-agentbrush-agent-brush
- Seller: https://agentstack.voostack.com/s/ultrathink-art
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.