Image Gen

Image Generation & Editing

Generate and edit images via OpenAI-compatible REST APIs using the bundled Python CLI script. No external dependencies required (Python 3.8+ standard library only). Compatible with OpenAI, Azure OpenAI (v1 endpoint), and other OpenAI-compatible providers.

Quick Reference

Supported Models

| Model | Generation | Editing | Notes | |---|---|---|---| | gpt-image-1 | ✓ | ✓ | Base GPT image model | | gpt-image-1-mini | ✓ | ✓ | Smaller, faster variant | | gpt-image-1.5 | ✓ | ✓ | Latest, most capable | | gpt-image-2 | ✓ | ✓ | Next-gen model (2026-04-21+) |

Key Parameters

| Parameter | Values | Default | Notes | |---|---|---|---| | size | auto, 1024x1024, 1536x1024 (landscape), 1024x1536 (portrait) | auto | | | quality | auto, low, medium, high | auto | | | background | auto, transparent, opaque | auto | Transparent requires png or webp output | | output_format | png, jpeg, webp | png | | | output_compression | 0–100 | 100 | Only for jpeg/webp | | n | 1–10 | 1 | Number of images | | moderation | auto, low | auto | Generation only |

Usage

The script at scripts/image_gen.py is a zero-dependency Python CLI.

Authentication

Variables can be defined in a .env file in the working directory or any parent directory.

| Variable | Description | Default | |---|---|---| | OPENAI_API_KEY | API key (required) | — | | OPENAI_BASE_URL | Base URL for custom endpoints (Azure OpenAI, etc.) | https://api.openai.com/v1 | | OPENAI_IMAGE_MODEL | Default model | gpt-image-1.5 |

All variables can also be passed as CLI flags (--api-key, --base-url, --model).

Generate an Image

python scripts/image_gen.py generate "A cat wearing a top hat" -o ./output

With full options:

python scripts/image_gen.py generate "A minimalist coffee shop logo" \
  --size 1024x1024 \
  --quality high \
  --background transparent \
  --output-format png \
  -o ./output

Edit an Image

python scripts/image_gen.py edit "Add sunglasses to the person" \
  -i photo.png -o ./output

Multiple input images (up to 16):

python scripts/image_gen.py edit "Create a gift basket with these items" \
  -i item1.png -i item2.png -i item3.png -o ./output

With mask for inpainting:

python scripts/image_gen.py edit "Replace the background with a beach" \
  -i photo.png --mask mask.png -o ./output

Important: Do not pass --model unless the user explicitly asks for a specific model. The default model is resolved from OPENAI_IMAGE_MODEL env var or falls back to gpt-image-1.5.

Prompt Guidelines

For best results with GPT image models:

• Be specific and descriptive: include subject, style, mood, colors, composition.
• Specify the art style: "photorealistic", "watercolor", "flat illustration", "3D render", etc.
• For logos/icons: request "transparent background" and set --background transparent --output-format png.
• For edits: clearly describe what should change and what should remain.
• The maximum prompt length for GPT image models is 32,000 characters.

Output

-o accepts a file path or a directory:

• File path (e.g. -o logo.png): saves directly to that path. With --n >1, appends _1, _2, etc. before the extension. Intermediate directories are created automatically.
• Directory (e.g. -o ./out): saves as image_1.{ext}, image_2.{ext}, etc.

Existing files are never overwritten — a numeric suffix is appended automatically if the target already exists.

The script prints saved file path(s) to stdout (one per line) and any revised prompts to stderr.

Recipes

Before generating an images, always first generate a prompt using these copy/paste instructions for prompt recipes: [references/sample-prompts.md](references/sample-prompts.md).

An image genaration/editing workflow consists of 2 steps:

1. Generate a prompt for the image generation/editing based on the user query using the sample prompts.
2. Pass the generated prompt to the image generation/editing command to generate/edit the image.

Source & license

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

• Author: sinedied
• Source: sinedied/agent-skills
• License: MIT

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