image-generation

Files (3)

• SKILL.md
• gemini.py · scripts/
• openai_image.py · scripts/

SKILL.md (first lines)

# Image Generation & Editing Skill

Generate and edit images using AI (Google Gemini Nano Banana Pro, OpenAI DALL-E 3).

**Capabilities:**
- 🎨 **Generate**: Create new images from text descriptions
- ✏️ **Edit**: Modify existing images (add/remove elements, change colors)
- 🛍️ **Product Placement**: Put products into scenes
- 🎭 **Style Transfer**: Apply artistic styles to photos
- 🖼️ **Composite**: Combine multiple images into one

## Quick Examples

Users can specify what they want:

| User Says | Mode | What Happens |
|-----------|------|--------------|
| "Generate an image of a sunset" | Generate | Text-to-image, no reference needed |
| "Create a logo for my coffee shop" | Generate | Text-to-image with text rendering |
| "Edit this image: add a hat to the cat" | Edit | User provides image, AI modifies it |
| "Remove the background from this photo" | Edit | User provides image, AI edits it |
| "Put this product on a kitchen counter" | Product | User provides product + optional scene |
| "Make this photo look like Van Gogh painted it" | Style | User provides photo, AI applies style |
| "Combine these photos into a group shot" | Composite | User provides multiple images |

## Prerequisites

Environment variables must be configured for the APIs to work. At least one API key is required:

- `OPENAI_API_KEY` - For OpenAI DALL-E 3 image generation
- `GOOGLE_API_KEY` - For Google Gemini (Nano Banana / Nano Banana Pro)

See the repository README for setup instructions.

## Available APIs

### OpenAI GPT Image (Recommended for pure generation)
- **Models**:
  - `gpt-image-1.5` (state of the art, best quality)
  - `gpt-image-1` (great quality, cost-effective)
  - `gpt-image-1-mini` (fastest, most affordable)
- **Best for**: High-quality generation, transparency, text rendering, image editing
- **Sizes**: 1024x1024 (square), 1536x1024 (landscape), 1024x1536 (portrait), or `auto`
- **Quality**: low (fast), medium (balanced), high (best), or `auto`
- **Background**: transparent, opaque, or `auto`
- **Output formats**: png (default), jpeg (faster), webp
- **Compression**: 0-100% (for jpeg/webp)
- **Features**:
  - Image editing with up to 16 input images
  - Transparent backgrounds
  - Streaming with partial images
  - High input fidelity for preserving faces/logos
  - Inpainting with masks
  - 32,000 character prompts

> ⚠️ **Note**: DALL-E 2 and DALL-E 3 are deprecated and will stop being supported on 05/12/2026.

### Google Gemini Native Image Generation (Recommended for editing)
- **Nano Banana** (`gemini-2.5-flash-image`): Fast, efficient, 1K resolution, up to 3 reference images
- **Nano Banana Pro** (`gemini-3-pro-image-preview`): Professional quality, up to 4K, thinking mode, up to 14 reference images (default)
- **Aspect ratios**: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9
- **Resolutions** (Pro only): 1K, 2K, 4K
- **Features**: 
  - Image editing (add/remove elements, color changes)
  - Product placement and composition
  - Style transfer
  - Advanced text rendering
  - Google Search grounding (Pro only)
  - Thinking mode for complex prompts (Pro only)

## Workflow

### Step 1: Gather Requirements (REQUIRED)

⚠️ **Use interactive questioning — ask ONE question at a time.**

#### Question Flow

⚠️ **Use the `AskUserQuestion` tool for each question below.** Do not just print questions in your response — use the tool to create interactive prompts with the options shown.

**Q0: Model Selection**
> "Which image generation model would you like to use?
>
> - Google Gemini (Nano Banana Pro) - Up to 4K, 14 reference images, style transfer, thinking mode (Recommended)
> - OpenAI GPT Image 1.5 - State of the art, transparency, streaming, up to 16 input images
> - OpenAI GPT Image 1 - Great quality, transparency, image editing
> - OpenAI GPT Image 1 Mini - Fastest, most affordable"

*Wait for response. If user doesn't have a preference, recommend Gemini for editing/reference tasks or GPT Image 1.5 for pure generation.*

**Q1: Reference**
> "I'll generate that image for you! First — **do you have any reference images?**
> 
> - Product photos to include
> - Style references
> - Images to edit
> - No, generate from scratch"

*Wait for response.*

**Q2: Aspect Ratio**
> "What **aspect ratio**?
> 
> - 1:1 (square)
> - 16:9 (landscape/widescreen)
> - 9:16 (portrait/vertical)
> - 4:3 / 3:4 (classic)
> - Other (2:3, 3:2, 4:5, 5:4, 21:9)
> - Or specify"

*Wait for response.*

**Q3: Resolution**
> "What **resolution**?
> 
> - 1K (fast)
> - 2K (balanced)
> - 4K (highest quality)"

From michaelboeding/skills, MIT.

More from michaelboeding/skills