Generate an image from a text prompt using OpenAI's image model family (models: "gpt-image-2.5-sunburst" (default), "gpt-image-2.5-flare", "gpt-image-2"). The image is written to disk and also returned inline so you can see it. These models handle photoreal, illustrations, infographics, multilingual text (incl. CJK), and complex structured visuals. For a transparent logo pass background="transparent" with output_format="png" — verified against this family; the origin decides, so check applied.background. Sizes accept presets or any custom "WxH" where edges are multiples of 16, max edge ≤ 3840px, aspect ratio within 1:3–3:1, and total pixels 655,360–8,294,400. Outputs above 2K are beta. Image work is slow, so this call hands off immediately: the first response carries a job_id — poll get_image_job until it reports state "completed" (the result arrives verbatim, images included). Set GPT_IMAGE_2_ASYNC_AFTER_MS to wait inline instead (below your host's tool-call timeout).

Edit or compose images with OpenAI's image model family (models: "gpt-image-2.5-sunburst" (default), "gpt-image-2.5-flare", "gpt-image-2"). Give 1–8 input images plus a text prompt; optionally include a PNG mask whose transparent regions mark what to change (mask applies to the first image). Great for: swap backgrounds, retouch products, combine multiple reference images into one composition, maintain a character across scenes. These models always process inputs at high fidelity (no input_fidelity knob needed). The edited image is saved to disk and returned inline. Image work is slow, so this call hands off immediately: the first response carries a job_id — poll get_image_job until it reports state "completed". Set GPT_IMAGE_2_ASYNC_AFTER_MS to wait inline instead (below your host's tool-call timeout).

Poll a backgrounded gpt-image job: every image tool (generate_image, edit_image, start_edit_session, continue_edit_session) hands off immediately and returns a job_id rather than blocking past MCP client timeouts. Call this with that job_id until state becomes "completed" (files are already written to disk and returned inline, with the model, requested/applied settings, usage, and cost) or "failed". While still "running", wait a few seconds between polls.

List the image jobs this server still remembers (newest first), with their state, prompt preview, and elapsed time. Useful after a reconnect or a context reset to recover a job_id instead of re-running — and paying for — a generation you already started. Running jobs are never expired; finished ones are kept up to GPT_IMAGE_2_JOB_MAX / GPT_IMAGE_2_JOB_TTL_MS and are lost on server restart.

Begin a stateful multi-turn edit session. Returns a session_id you then pass to continue_edit_session to iteratively refine the image (each turn uses the previous turn's output as the input). Use end_edit_session when done. The first turn hands off to a background job like every other image call: poll get_image_job for it, and the session_id arrives with its "completed" result.

Apply another edit turn to an existing session. The previous turn's output image is used as the input. Use short, focused prompts like "make the sky more orange" or "add a small boat on the horizon"; include "keep everything else the same" to limit drift. Returns the new image and the updated session. Omitting size, quality, background, output_format, or model inherits what the session already uses (list_edit_sessions reports those settings). The turn hands off to a background job (poll get_image_job); while one is still running the session stays busy and further turns are refused until it lands.

Free an iterative-edit session. Safe to skip — sessions are in-memory only and are discarded on server restart — but calling this frees memory sooner and keeps list_edit_sessions tidy.

List active iterative-edit sessions (in-memory only, discarded on server restart). Useful to recover a session_id after a client reconnect; each entry also reports whether a turn is still running in the background and which job to poll for it.