Generate an Image via BlockEden OpenAI Proxy

Example: text-only generation

python /path/to/skill/scripts/generate_image.py \
  "a cozy coffee shop on a rainy day, impressionist painting style" \
  --size 1792x1024 \
  --quality hd \
  --style natural \
  --output-dir ~/Desktop

Example: reference image generation

python /path/to/skill/scripts/generate_image.py \
  "same composition but set at night with neon lights" \
  --reference-image "https://example.com/my-photo.jpg" \
  --output-dir ~/Desktop

After generation

Text-only mode output

{
  "filepath": "/path/to/image_1234567890.png",
  "revised_prompt": "...",
  "model": "dall-e-3",
  "size": "1024x1024",
  "quality": "standard",
  "style": "vivid"
}

Reference image mode output

{
  "filepath": "/path/to/image_1234567890.png",
  "model": "gpt-image-1",
  "reference_image": "https://example.com/my-photo.jpg",
  "prompt": "same composition but set at night with neon lights"
}

Once the image is saved:

1. Tell the user the file path so they can open it.
2. Show the revised_prompt if DALL-E rewrote it — this helps the user refine future prompts.
3. Offer to regenerate with different parameters if the user wants to iterate.

Interpreting the user's request

• Extract the core visual description from what the user says and use it as the prompt.
• If the user provides a reference image URL/URI, pass it via --reference-image. The model switches automatically to gpt-image-1.
• If the user specifies an aspect ratio (wide, tall, square), map it to the appropriate --size.
• If the user says "high quality", "detailed", or "HD", add --quality hd (text-only mode only).
• If the user says "realistic" or "photorealistic", use --style natural.
• If the user says "artistic", "dramatic", or "vivid", use --style vivid (the default).
• If the user provides an output location, pass it via --output-dir.
• If no output location is given, save to the current working directory.

Prompt crafting tips

Good image prompts are specific and visual. If the user's request is vague, enhance it slightly:

• Add a medium/style if none given (e.g., "digital art", "watercolor", "photorealistic")
• Add lighting or atmosphere if it feels generic (e.g., "golden hour lighting", "moody atmosphere")
• Don't over-engineer — keep the user's intent central