Copilot Can Generate Images in Your Project—Treat the Result Like Any Other Asset

A small visual can interrupt a coding task more than its file size suggests. You leave the editor, describe the same concept in another tool, download a result, find the right project folder and then return to the code. GitHub’s October 28 preview tries to collapse that loop: ask Copilot for an image, inspect the result inline and save it into the workspace. The convenience is real, but so is the need to review the artifact before it ships.

From chat request to project file

GitHub says the feature is available in Copilot chat in Visual Studio Code, the Copilot app and Copilot CLI. A selected reasoning model can call a dedicated image-generation tool powered by GPT Image 2.5 Flare. Copilot previews the output inline and saves it directly to the project. The preview is available for Copilot Pro, Pro+, Business and Enterprise plans, with image tokens billed separately and counted toward plan AI credits. ^[1]

OpenAI’s model documentation describes GPT Image 2.5 Flare as its faster high-quality image model, and its public release explains the model family’s improvements in image quality, editing and speed. Those are model-provider claims, not a guarantee about the image a particular prompt will produce or how well that image fits a codebase. The relevant product change is the integrated workflow around the model: the developer can stay in context while generating and saving a file. ^{[3] [4]}

This is most useful for provisional assets: a concept illustration to test page composition, a placeholder that is intentionally marked for replacement, or a small custom visual that would otherwise require a costly context switch. It is less useful when the asset has exact brand, legal, identity or data requirements that a generated image cannot establish by itself.

The saved file still belongs to the software supply chain

Once an image sits in a repository, the same product questions apply as for any other asset. Is it the right crop at mobile card size? Does its alt text describe the visible image rather than the prompt? Is it compressed appropriately? Does it accidentally contain malformed lettering, a recognizable mark, a misleading pseudo-interface or visual details that look like evidence?

A team should also decide where generated assets are allowed. GitHub says organization and enterprise owners can turn off image generation by disabling the image-generation model through Copilot policies. Its policy documentation describes controls over feature and model availability, making an organization-level decision possible instead of leaving every developer to infer the rules. ^{[1] [2]}

The changelog says a visible “Made with AI” watermark can be enabled, but is off by default. That setting does not replace a project’s own provenance record. If generated art is used in a public product, teams may want to record the tool, prompt, date, human edits and review decision alongside their existing design process. The specific recordkeeping obligation depends on the organization and use case; the product announcement does not define a universal disclosure rule. ^[1]

Keep creative iteration separate from production approval

A practical pattern is to let a developer ask for several directions, then move only the selected asset into a normal review. Keep the generated output in an intentional folder, compare it against the design system, test its crop and contrast at actual display sizes, and inspect the image for stray text or visual inconsistencies. Have a human author meaningful alt text and ensure the image does not make a claim that the surrounding article cannot support.

The workflow can also benefit from reproducibility. Save the prompt and source image when the asset is expected to be revised later, and avoid overwriting approved artwork by default. A small file-level diff is easier to review than an unexplained binary replacement. For broader agent workflows, see our guide to Copilot’s model-routing system and the recent discussion of Copilot’s Xcode integration.

A shorter loop, with the same judgment

Putting image generation beside code is a product-design choice: it makes a visual output reachable at the moment the developer discovers a need. That can make prototypes faster and reduce tool switching. It also makes it easier for an unreviewed asset to become part of a build. The best adoption keeps the creative step lightweight and the shipping step deliberate. A saved image is an artifact; it becomes a product asset only after someone confirms it belongs there.