OpenAI sets an API shutdown date for GPT Image 2

Imagine an online shop with a button that replaces a product photo’s background. The shop app sends the photo and editing instructions through an API, the connection another program uses to ask OpenAI for work. It also sends a model name. The returned image goes back to the shop for review. An API retirement affects that request even if the button in the shop’s app still looks the same.

%% How a fictional shop requests an edit
flowchart TD
A["Staff: choose a photo"] --> B["App: request an edit<br/>with a model name"]
B --> C["Model: return an image"]
C --> D["Staff: check the result"]

A retirement notice gives the app team time to test a replacement. Once the old names are removed, the app must already have a supported model configured to keep sending new requests. A dated snapshot is a fixed model version, and the notice explicitly includes that snapshot too.

Suppose the fictional shop wants a white background while keeping the volume printed on the product label. Give both models the same photo, editing instructions, size, format and quality setting. Look at whether the label still says what it should and whether the image can pass the shop’s approval process. Record response time and actual API usage on the same basis. A preview that looks good can still fail the label check.

OpenAI’s guide says GPT Image models may require API organization verification. Check access using the organization that will actually run the app. Measure usage before estimating cost. The shop example illustrates a test design and contains no measured quality, latency or cost. Image generation guide