How to Enable Local Image Generation for the gpt-image-2 Skill

Set ENABLE_GARDEN_IMAGEGEN=1 and provide a valid OPENAI_API_KEY to activate Mode A, then execute the generation scripts to create PNG files locally in the garden-gpt-image-2/image/ directory.

The gpt-image-2 skill in the ConardLi/garden-skills repository supports three distinct runtime configurations, with Mode A enabling full local image generation using OpenAI's GPT-Image-2 model. To enable local image generation for the gpt-image-2 skill, you must configure specific environment variables that signal the runtime to execute the local generation pipeline rather than delegating operations to external hosts or operating in advisory-only mode.

Understanding the Three Runtime Modes

The skill determines its operational mode at startup by inspecting environment variables in skills/gpt-image-2/scripts/check-mode.js:

• Mode A (Garden Local): Executes generate.js and edit.js locally to call OpenAI's /images/generations and /images/edits endpoints, saving outputs to the local filesystem.
• Mode B (Host-Native): Delegates prompt generation to the host environment without executing local generation scripts.
• Mode C (Advisor): Functions purely as a prompt advisor without producing image files.

Only Mode A performs local image generation. When properly configured, check-mode.js outputs:

--- gpt-image-2 runtime mode ---
mode = A
recommendation = Use local generation (generate.js / edit.js)

Configuration Prerequisites

Two environment variables must be configured to enable local generation:

• ENABLE_GARDEN_IMAGEGEN: Must be set to 1, true, yes, or on to signal Mode A activation.
• OPENAI_API_KEY: Must contain a valid OpenAI API key for authenticating requests to the image generation endpoints.

Create a .env file in the repository root with these values:

ENABLE_GARDEN_IMAGEGEN=1
OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXXXXXX

Step-by-Step Enablement

Follow these steps to activate and verify local image generation:

1. Configure environment variables in .env or export them directly in your shell.
2. Verify Mode A detection by running the mode checker.
3. Prepare directory structure (garden-gpt-image-2/prompt/ and garden-gpt-image-2/image/).
4. Execute generation using the provided scripts.

Verifying Your Configuration

Before generating images, confirm the skill has detected Mode A by executing skills/gpt-image-2/scripts/check-mode.js:

node skills/gpt-image-2/scripts/check-mode.js --json

Expected output for local generation mode:

{
  "mode": "A",
  "recommendation": "Use local generation (generate.js / edit.js)"
}

If the output shows mode: "B" or mode: "C", verify that ENABLE_GARDEN_IMAGEGEN is set to a truthy value and that the environment variables are loaded into your shell session.

Generating Images Locally

Once Mode A is active, the generation pipeline executes as follows:

1. Select a template and render a prompt.
2. Save the prompt to garden-gpt-image-2/prompt/.
3. Invoke generate.js, which calls POST /images/generations on the OpenAI API.
4. Store the resulting PNG under garden-gpt-image-2/image/.

Default directories are defined in skills/gpt-image-2/scripts/shared.js:

export const DEFAULT_IMAGE_DIR = "garden-gpt-image-2/image";
export const DEFAULT_PROMPT_DIR = "garden-gpt-image-2/prompt";
export const DEFAULT_MODEL = "gpt-image-2";

Generate an image from a prompt file:

node skills/gpt-image-2/scripts/generate.js \
  --promptfile garden-gpt-image-2/prompt/my-poster-20260902-101500.md

Alternatively, generate from a template directly:

node skills/gpt-image-2/scripts/generate.js \
  --template "ui-mockups/live-commerce-ui" \
  --output garden-gpt-image-2/prompt/live-commerce-ui-$(date +%Y%m%d-%H%M%S).md

To edit existing images rather than generating from scratch, use the edit script:

node skills/gpt-image-2/scripts/edit.js \
  --promptfile garden-gpt-image-2/prompt/edit-request.md

Key Implementation Files

The following source files handle local image generation in Mode A:

• skills/gpt-image-2/scripts/check-mode.js: Detects runtime mode based on environment variables.
• skills/gpt-image-2/scripts/generate.js: Entry point for image generation; calls OpenAI API and saves PNG outputs.
• skills/gpt-image-2/scripts/edit.js: Handles image editing via OpenAI's edits endpoint.
• skills/gpt-image-2/scripts/shared.js: Exports default directory constants and model configuration.
• skills/gpt-image-2/SKILL.md: Contains high-level skill documentation and mode specifications.

Summary

• Set ENABLE_GARDEN_IMAGEGEN=1 to activate Mode A for local generation.
• Provide OPENAI_API_KEY to authenticate with OpenAI's image API.
• Verify mode using check-mode.js before executing generation scripts.
• Execute generate.js to create images, which saves outputs to garden-gpt-image-2/image/.
• Use edit.js for modifying existing images rather than generating new ones.

Frequently Asked Questions

What happens if I don't set ENABLE_GARDEN_IMAGEGEN?

Without this variable set to a truthy value, the skill defaults to Mode B (host-native delegation) or Mode C (advisor-only), neither of which produces local image files. The generation scripts will not execute locally, and no PNG files are saved to the garden-gpt-image-2/image/ directory.

Can I use local image generation without an OpenAI API key?

No. Mode A requires a valid OPENAI_API_KEY because skills/gpt-image-2/scripts/generate.js makes direct HTTP requests to https://api.openai.com/v1/images/generations. Without authentication, the API calls fail and image generation cannot complete.

Where exactly are generated images saved?

According to skills/gpt-image-2/scripts/shared.js, the default output directory is garden-gpt-image-2/image/. Each successful generation creates a PNG file in this location with a timestamp-based filename, while corresponding prompts are stored in garden-gpt-image-2/prompt/.

How do I switch from generation mode to editing mode?

Use skills/gpt-image-2/scripts/edit.js instead of generate.js. Both scripts require the same environment configuration (ENABLE_GARDEN_IMAGEGEN=1 and OPENAI_API_KEY), but edit.js calls the /images/edits endpoint to modify existing images based on provided mask and prompt files.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}