Codex + Images: Instant PRs
- OpenAI integrated Images 2.0 into the Codex dev workspace so designers can generate and refine UI ideas in-app. - A demo from OpenAI’s design lead showed riffing UI concepts and turning that work into a pull request instantly. - The posts illustrate a one-flow: Codex image gen to UI spec to code, shortening prototype-to-PR cycles for product teams ([] []).
OpenAI has folded its new image generator into Codex, turning the coding app into a place where a design mockup can move toward a pull request in one workspace. (openai.com) OpenAI said on April 16 that the updated Codex app can generate images, use an in-app browser, review pull requests, run multiple terminals, and connect to remote development boxes over Secure Shell, or SSH, in alpha. The company said more than 3 million developers use Codex every week. (openai.com) Codex started as a cloud coding agent in May 2025 that could read and edit a repository, run tests in an isolated environment, and then let a user open a GitHub pull request or merge the changes locally. OpenAI later introduced a desktop Codex app on February 2, 2026, and expanded it to Windows on March 4. (openai.com 1) (openai.com 2) Image generation is the new piece in that chain. OpenAI said on April 21 that ChatGPT Images 2.0 improves text rendering, multilingual output, and visual reasoning, which are the parts of image models that matter when a team is mocking up buttons, menus, labels, and screen layouts. (openai.com) That combination changes what “prototype” means inside a coding tool. Instead of handing off a static mockup from one app to a designer, a specification in another app to a product manager, and implementation notes to an engineer, Codex now bundles screenshots, generated images, browser feedback, code edits, and GitHub review into one thread. (openai.com) (developers.openai.com) OpenAI’s own product notes describe that shift in practical terms. The April 2026 Codex changelog says the app can preview generated artifacts in the sidebar, comment directly on rendered pages in the browser, and inspect GitHub pull requests and review comments without leaving the app. (developers.openai.com) The company has been widening Codex beyond code generation for months. In the February app launch, OpenAI described Codex as a “command center” for multiple agents working in parallel on the same repository through isolated worktrees, with users able to review diffs, comment on changes, and open the result in an editor. (openai.com) OpenAI has not published a separate product post for the social media demo itself, and the two X links attached to this story did not return readable page text in web fetches on April 23. But the workflow shown in those posts matches the capabilities OpenAI documented across the April 16 Codex update, the April 20 changelog entry, and the April 21 Images 2.0 release. (openai.com) (developers.openai.com) (openai.com) The immediate effect is less about replacing design tools than collapsing steps between idea, mockup, implementation, and review. OpenAI’s product pages now describe Codex as software that can generate the visual concept, edit the code, run checks, and carry the work into a pull request from the same workspace. (openai.com 1) (openai.com 2)
