AustineA/app-landing-seo

GPT Image

Generate images with OpenAI's gpt-image-2 Images API through this skill's bundled helper.

Setup

• Prefer uv; the script declares its openai dependency inline.
• Otherwise install openai with python3 -m pip install -r requirements.txt from this skill directory.
• Require GPT_API_KEY or OPENAI_API_KEY in the environment.
• Resolve GPT_IMAGE_SKILL_DIR to the directory containing this SKILL.md. Run the helper as "$GPT_IMAGE_SKILL_DIR/scripts/generate_image.py" so it works from any project directory.

Generate

"$GPT_IMAGE_SKILL_DIR/scripts/generate_image.py" "<detailed prompt>" \
  --aspect-ratio 3:2 \
  --quality high \
  --background opaque \
  --format webp \
  --output "<destination>.webp"

Options:

• --aspect-ratio: 1:1, 16:9, 9:16, 3:2, 2:3, auto, or an explicit supported size.
• --quality: low, medium, high, or auto.
• --background: auto, transparent, or opaque. Transparency requires PNG or WebP.
• --format: png, jpeg, or webp.
• --reference PATH: condition on an existing image; repeat for multiple references.
• --sprite: optimize the prompt for a centered game asset and default to a transparent background.
• --model: defaults to gpt-image-2; keep that default unless the user requests another compatible model.

Prompt and output checks

Describe the subject, composition, viewpoint, lighting, materials, palette, and exclusions that matter. For feature images, explicitly prohibit text, letters, numbers, captions, watermarks, interface labels, and logos unless they are intentional.

Inspect every generated image at full size before using it. Check requested dimensions and format, subject relevance, crop safety, malformed objects, accidental glyphs, unintended branding, and any mismatch with reference images. Regenerate only when a concrete defect is visible, and change the prompt to address that defect rather than repeating the same request.

The helper creates missing output directories. It does not automatically open generated files.