Codex image gen goes default
- Codex's workspace/CLI image generator was flipped on as a default option with queue‑based generation. - New commands include codex --enable image_generation and the $imagegen shortcut, paired with GPT‑Image‑2. - Developers say the change improves high‑detail handling and speeds prototype iterations in the workspace ( ).
OpenAI has turned image generation on by default in Codex, folding visual asset creation directly into the coding agent’s workspace and command-line tool. (developers.openai.com) Codex is OpenAI’s coding agent for local development: it runs in a terminal, reads a repository, edits files, and executes commands in the selected directory. OpenAI’s current documentation says the tool can now “generate or edit images directly in the CLI” alongside normal code work. (help.openai.com) The command-line reference shows Codex supports feature flags through `--enable` and `--disable`, which map to settings in `~/.codex/config.toml`. OpenAI’s developers said this image feature can be toggled with `codex --enable image_generation`, and they also pointed users to an `$imagegen` shortcut inside the workspace. (developers.openai.com, x.com) OpenAI introduced `gpt-image-2` on April 21, 2026 and said the model is available both in the API and in Codex. In that launch note, the company described it as its most capable image model for “complex visual tasks,” with stronger editing, layout control, text rendering, and instruction-following. (community.openai.com) Image generation in a coding agent solves a practical development problem: developers often need icons, mockups, diagrams, screenshots, and marketing art while they are still building the product. OpenAI’s Codex docs now place image generation beside code review, subagents, and image-input support as part of the same terminal workflow. (developers.openai.com, help.openai.com) The model launch also pushed beyond simple illustrations. OpenAI said `gpt-image-2` supports more aspect ratios, resolutions up to 2K, and text-heavy outputs such as diagrams, infographics, charts, posters, comics, and multilingual assets. (community.openai.com) Codex already accepted image inputs before this change, which let developers attach screenshots or design specs for the agent to read. The new default means the same session can now move from reading a mockup to producing one, instead of handing the job off to a separate design tool or API script. (help.openai.com, developers.openai.com) OpenAI’s developers said the queue-based setup is meant to keep work moving while image requests run in the background, and said the updated flow handles high-detail generations better inside the workspace. That puts visual iteration closer to the same loop developers already use for prompts, diffs, approvals, and follow-up commands in Codex. (x.com, developers.openai.com)
