MCP AI Image Guide

AI Image through MCP

AI Image uses image. MCP exposes model settings as flat kp_create_creation arguments rather than the OpenAPI image_generation object.

Always call kp_get_catalog before choosing a model or ratio.

Nano Banana 2

{  "type": "image",  "prompt": "A clean course cover about binary search in a dark vector style",  "aspect_ratio": "16:9",  "output_quality": "2K",  "image_provider": "gemini",  "image_model": "nano-banana-2"}

GPT Image 2

{  "type": "image",  "prompt": "A precise binary search poster for a developer course",  "aspect_ratio": "3:2",  "output_quality": "2K",  "image_provider": "azure_openai",  "image_model": "gpt-image-2",  "image_quality": "medium"}

image_quality is the model-native low / medium / high setting. It is separate from the top-level final output specification output_quality.

When to choose AI Image

Use AI Image for a standalone illustration, cover, diagram-like composition, or visual reference. Describe the subject, purpose, composition, and style in the prompt. If exact wording, a dense table, or a legally significant label matters, review the result before using it; image generation is not a substitute for source verification.

Select settings from the catalog

kp_get_catalog determines which provider, model, aspect ratio, and quality combinations are currently accepted. Keep the top-level output_quality separate from image_quality: the first describes the requested result, while the second is the GPT Image model setting. Do not combine a ratio or a quality value from one model with another provider's model.

After kp_create_creation, poll the returned job with kp_get_job_status. On success, use kp_get_creation to retrieve the actual main_url and artifacts. For an error, change the unsupported catalog value or clarify the request before trying again.

Continue from here

Quickstart for authentication and the job lifecycle
AI Video for time-based scenes with native audio
Available Tools for model and capability discovery