Image to Code skill for Codex.
The Image to Code skill has Codex rebuild a selected image, screenshot, mockup or Image Gen result as a faithful, responsive and interactive frontend. It ships in OpenAI’s Product Design plugin and won’t start from a written brief alone.
Updated November 12, 2026 · 3 min read
At a glance
| Skill name | image-to-code |
| Comes with | The Product Design plugin, by OpenAI |
| Install | codex plugin add product-design@openai-curated, or /plugins in Codex |
| Codex uses it | When you’ve picked an image, screenshot, mockup or generated design and want it built as a working frontend |
| Source | openai/plugins on GitHub |
What it does
The skill is written as a checklist, not loose guidance. First it pins down the exact image to build. If you chose a numbered option from the ideate skill, it maps that number to the images in the order they were shown, and it asks you rather than guessing when the choice is unclear.
Then it works through the build:
- Assets: it lists every image in the design (hero art, thumbnails, logos, avatars, textures) and generates each one with Codex’s Image Gen tool. It is told never to fake them with CSS shapes, hand-drawn SVGs or emoji.
- Layout: it measures each section’s spacing and sizes, finds free fonts that match, and picks the closest icon library rather than defaulting to Lucide.
- Behavior: navigation, tabs, menus, forms, filters and hover, loading, empty and success states work. Auth, backends and persistence are left out unless you ask.
- Mobile: a mobile-sized design becomes a mobile app built on the plugin’s
mobile-apptemplate, with its own device frame and status bar.
When the app runs, Codex screenshots it and runs the design-qa skill against the original image. It fixes every P0 to P2 issue and repeats until design-qa.md in the project root says final result: passed.
When to use it
Use it when you already have the look you want, from a screenshot, a mockup, a Figma frame or an ideate option, and want a prototype that matches it closely. If you only have an idea, start with ideate to get images first.
Try it
Use the image-to-code skill. Build the attached mockup as a responsive, interactive web app. Generate the images it shows, make the tabs and search work, and run design QA against the mockup before you hand off.
Does it work in Claude Code?
Only in part. The skill depends on Codex-only pieces: the Image Gen tool for every asset, the Codex in-app browser for screenshots, and other files in the Product Design plugin (its router skill, the design-qa skill and the prototype bootstrap script). Copying this one SKILL.md folder into ~/.claude/skills leaves those links broken, so Claude Code can follow the checklist only if you supply images and handle screenshots another way.