GPT Image2 High-Quality AI Image Generation

Problem

gpt-image-2 can take about 150 seconds per image, so synchronous calls easily hit common 60s timeouts. Batch generation, reference images, and text rendering also require reliable job tracking and polling. This skill packages async submission, result polling, login credentials, and rate handling into command-line tools, allowing an agent to generate images without manually managing job_id or wait loops.

Core Workflow

• Submit: call gpt-image2.gpt_image2_submit with prompt, size, n, style, and reference_image_urls, then receive a job_id.
• Poll: call gpt-image2.gpt_image2_result; the server already waits 50 seconds, so the client should use --timeout 70 until the status becomes done or error.
• Style control: 20 built-in style presets cover Ghibli, Pixar, claymation, LEGO, cyberpunk, botanical engraving, and more; reference image URLs can support image-to-image generation and style transfer while preserving subjects.

Boundaries

• Every submit consumes real credits; batch n > 1 or retries should be treated as separate charges.
• Prompts and reference URLs are sent through the gateway, so confidential or sensitive personal data should not be included.
• If a job is still running after 600 seconds, keep the job_id and continue polling instead of declaring failure immediately.