What Polling Checks While an Image Task Is Open
A function that returns its result before it ends is the pattern in a first programming course. This image route is a different pattern. The request that starts the work and the request that collects the picture are not the same call. Between them, the client has to keep a task id and ask again. That second ask is polling. The status it reads is the part to understand before treating the call as a single line that comes back with pixels.
The call the client is making
The GPT Image 2.5 Sunburst API is Topview’s route to OpenAI’s GPT Image 2.5 Sunburst. The route name on the page is openai/gpt-image-2.5-sunburst. Two modes are listed: text to image, and image editing. Image editing takes one input image together with a prompt. The prompt can run to 4,000 characters. Quality is low, medium, high, xhigh, or max. Medium is the default. Resolution is 1K, 2K, or 4K. 1K is the default. The frame is 1:1, 16:9, or 9:16. Price is per generated image, and it changes with those two settings. The page names the pattern in one line: asynchronous submission, then polling, then output retrieval.
What you have after the first response
The first request submits a prediction. It carries the API key and an idempotency key. The field the page tells you to save is the task id. The image comes later, when a poll of that id says the task has completed. A program that prints the first body and exits has opened a job and dropped the handle it needed for the next request. The page’s sample body is a prompt for a blue ceramic cup on a wooden table, with aspect ratio 1:1, resolution 1k, and quality medium. That sample belongs to the docs. It is not a picture produced for this lesson.
What each poll is reading
Polling is a later request that names the same task id and reads the status. The page says to continue until the task is completed or failed. Completed is the point where you retrieve the image URLs and the usage cost. Failed is a stop. It is not an instruction to poll the same id without looking at the status. The page does not publish a required pause between polls, so this lesson does not invent one. It does say to download the images promptly, because result URLs can expire. A URL left only in a notebook is not a saved file.
Where the key belongs
The API key is how the platform knows the caller. The page says to keep it on the server. A page of browser script, a screenshot in a class chat, or a repository a classmate can clone is the wrong place for it. The image can still be shown to the class after the server has downloaded it. The key does not have to travel with the picture. Keep the same idempotency key, and the same input, only when you are retrying a request you are not sure arrived. The page’s note treats a new task as a new key. Another prompt, another quality, or another resolution is a new task, even when the sentence in the prompt barely changed.
What a completed image charges
API use is paid from the Platform USD balance. Promotional credits vary, and the page says to check the current balance. The Playground is where you inspect settings. The control that offers a free try copies the prompt into that Playground, and you confirm the settings there. That step is not a daily free allowance of API calls. On the text-to-image table, the default row is medium quality at 1K. Topview lists 0.0120 US dollars per image. The official rate on that row is 0.0240 US dollars, which the page calls 50 percent lower. Image editing costs slightly more than text to image at the same setting. The page holds the rest of the grid, including other platforms whose quotes may not match quality or dimensions. This lesson does not copy those rows. Read the row for the quality and resolution you actually send. GPT Image 2.5 Flare shares the modes, the parameters, and the current rates on this platform. The page says to run the same prompt in each Playground. Matching settings do not decide which picture is better. The route in this lesson is Sunburst.
What the file means after the status changes
Text to image makes a new picture from the prompt. Image editing asks for a change to one image you have the right to upload. The page describes composition, color, and styling, along with moodboards and poster-like key art. Samples shown there include a film poster, a card-game screen, a pixel sheet, and a manga page. Those examples are the page’s. The result is a new picture. Small type inside it can be wrong. Do not hand it in as a photograph of a lab, a campus event, or a person who was not there. This route does not take a video, and it does not remove an object from a moving shot.
The sequence a student can trace
Name the mode, the prompt, the ratio, the resolution, and the quality before the first request. Submit that prediction from the server. Store the task id the response gives you. Poll that id until the status is completed or failed. If the first request timed out and you never saw a response, repeat it with the same idempotency key and the same input. If you want a different picture, start a new task with a new key. When the status is completed, download the image and read the cost. When it is failed, stop on that id. A new attempt is a new task, not another unread poll.
What you can say from the page alone
The first request and the poll are different calls, joined by the task id. The key stays on the server. Editing accepts one image. Medium and 1K are the defaults, and the default text-to-image row on the page is 0.0120 US dollars per image. A result URL is for downloading now. It is not a promise that the same address will still open later.
