Skip to content
NewsProduct launch

GitHub Copilot Can Now Generate and Edit Images Directly in Chat

· by Pondero Newsdesk

The short version

GitHub shipped a public-preview image-generation and editing tool inside Copilot chat on October 28, powered by a model called GPT Image 2.5 Flare and billed separately through AI credits.

GitHub Copilot Can Now Generate and Edit Images Directly in Chat

Asking Copilot for a diagram or a mockup used to mean leaving the chat, describing the same request to a separate image tool, and copying the result back into the project. GitHub removed that step on October 28, shipping a public-preview tool that lets Copilot generate and edit images inline, using a dedicated model called GPT Image 2.5 Flare, per GitHub's changelog.

What shipped

A developer can describe an image in plain language inside any Copilot chat session, whether starting a fresh request or mid-task on something else, and Copilot calls the image model, previews the result inline, and saves it to the project workspace, according to the changelog. The feature works the same way across Copilot CLI, the GitHub Copilot app, and Copilot chat in VS Code. It is live in public preview for Copilot Pro, Copilot Pro+, Copilot Business, and Copilot Enterprise. GitHub also added an opt-in visible "Made with AI" watermark on generated and edited images; it stays off unless a user or org turns it on.

Billing runs on a separate track from the chat model a developer picked. Image requests draw on "image tokens" that count against the AI credits attached to a Copilot plan rather than the per-message quota of whichever reasoning model is handling the conversation, the changelog states. Organization and enterprise owners can shut the capability off entirely through Copilot model policies, the same admin surface GitHub already uses to restrict individual models account-wide, per GitHub's documentation.

Image generation now draws from the same credit pool as everything else

Teams that treated their Copilot AI-credit allotment as a coding-only budget need to recalculate once this leaves preview. A design review that used to cost nothing beyond a developer's time, sketch a few icon variants, mock up a UI state, now consumes credits the same way a long agent run does. For a Business or Enterprise admin, the practical decision this week is whether to leave image generation on by default or require it be approved, because the policy toggle exists but nothing in the preview enables it off by default. Smaller teams on a Pro or Pro+ seat should watch their credit burn rate for a few weeks before leaning on the feature for anything beyond occasional use.

The watermark default is worth a second look too. GitHub ships "Made with AI" as opt-in, not opt-on, which means an image generated inside Copilot and committed straight into a public repo or shipped in a product carries no visible marker unless someone turns that setting on. Teams with disclosure obligations around AI-generated creative assets, marketing material pulled from a repo, documentation screenshots, client-facing mockups, will need to set that policy explicitly rather than assume it is already covered.

Context and reactions

The image tool landed as one piece of a broader set of releases GitHub pushed out around GitHub Universe 2026 on October 28, alongside separate changes to how Copilot routes between models and how it attaches structured artifacts to issues. GitHub frames image generation as closing a context switch rather than competing with standalone image generators: the changelog's own framing is that getting a visual asset previously meant stepping outside Copilot, describing the same idea again elsewhere, and bringing the result back in by hand.

What to watch next

Watch for GitHub's first usage or cost guidance once enough preview users have burned through real credit budgets, since the company has not published example image-token pricing the way it has for chat-model tokens. Also watch whether the "Made with AI" watermark becomes a per-organization default rather than a manual opt-in, which would be the more defensible setting for teams with external disclosure requirements.

Sources