Generated images are not saved to disk and no path is returned (regression in 0.140/0.141, Linux)

Open 💬 3 comments Opened Jun 20, 2026 by Vulxhisers

What version of Codex CLI is running?

codex-cli 0.141.0 (broken on 0.140.0 and 0.141.0; last working: 0.139.0)

What subscription do you have?

ChatGPT Pro

Which model were you using?

gpt-5.5

What platform is your computer?

Ubuntu 24.04.4 LTS (noble), kernel 6.17.0-35-generic x86_64, Node v24.6.0

What terminal emulator and version are you using (if applicable)?

GNOME Terminal (default Ubuntu terminal)

Codex doctor report

not available

What issue are you seeing?

Built-in image generation no longer writes a file to disk and returns no path, starting with 0.140.0 (also broken on 0.141.0). On 0.139.0 the same prompt generated an image, saved it under ~/.codex/generated_images/, and printed the saved path in the console.

On 0.140.0 and 0.141.0 the agent reports that it generated and saved the image, but:

  • no PNG/JPG appears in ~/.codex/generated_images/
  • no saved_path/file_path is printed or returned to the agent (path = null)
  • the file does not exist anywhere on disk

I verified this repeatedly on both 0.140.0 and 0.141.0 — the file is definitely never created. After downgrading to 0.139.0 it works correctly again (file is saved and the path is printed).

What steps can reproduce the bug?

  1. Install codex 0.140.0 or 0.141.0 on Ubuntu 24.04.
  2. Ask it to generate an image, e.g. "generate a red square and save it".
  3. The agent claims the image was generated and saved.
  4. Check ~/.codex/generated_images/ → empty. No path printed. No file on disk.

Downgrading to 0.139.0 makes the same steps work as expected.

What is the expected behavior?

image_gen should write the file to disk (e.g. under ~/.codex/generated_images/) and return/print a usable saved path, as it did in 0.139.0.

Additional information

Regression range:

  • Working: 0.139.0
  • Broken: 0.140.0, 0.141.0 (verified on both)

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗