Copilot can make the image while you build the page

A missing hero image can interrupt a page build as surely as missing code. GitHub’s October 28 public preview lets Copilot generate or edit that asset in the coding conversation, preview it inline and save it into the project workspace. GitHub’s release.

The selected reasoning model calls a separate image tool powered by GPT Image 2.5 Flare. That distinction matters when comparing results or costs: changing the model that discusses the page does not make the image request the same kind of work.

From a preview to an asset that fits

GitHub lists VS Code, the Copilot app and Copilot CLI, with preview access for Pro, Pro+, Business and Enterprise. Organization and enterprise owners can disable the image model through Copilot model policies.

For a proposed page-building task, describe the space the image must occupy: a wide crop, a subject to one side and room beside it for the headline. Then inspect the saved file at the actual desktop and phone sizes. An attractive chat preview can still crop the subject out on a narrow screen. Check any visible lettering against the required copy rather than assuming it is correct. This is a suggested review, not an image test we performed.

Image usage has its own charge

The release says image tokens consume plan AI credits separately from the chat model. It does not give a fixed per-image price, so “one image” should not become an invented budget unit. Inspect the applicable image usage charge before asking for a large set of alternatives.

A visible “Made with AI” watermark is optional and off by default, according to GitHub. Decide how the finished page will explain an illustrative image’s role, especially if readers might mistake it for a product screenshot or measured result.

Start with one asset and place it in the real layout before requesting more. Our AI output evaluation guide offers checks for the result; the Copilot model-policy story explains how inherited access can affect the tool you are allowed to use.