Built-in image_gen tool should accept explicit parameters: size, quality, output_format, filename/path

[chrm52]

What variant of Codex are you using?

CLI

What feature would you like to see?

The current built-in image_gen tool (enabled via [features] image_generation = true) only accepts a single prompt parameter.

This forces users to cram everything into natural language, which is unreliable for:

• Exact dimensions
• Output filename / save path
• Quality settings
• Format (PNG with real transparency, WebP, etc.)

Current behavior (CLI / App)

Generate a hero banner, save as assets/hero.png

### Additional information

_No response_

[github-actions]

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• Agent image-gen tool: no reference image input / no way to reproduce or iterate #19864

Powered by Codex Action

[chrm52]

Additional real-world scenario (why explicit filename is critical)

This becomes especially painful when multiple Codex agents (or parallel tool calls) are generating images at the same time.

What happens today:

• Agent A prompts: “generate hero banner for dashboard”
• Agent B prompts: “generate icon for settings page”
• Both run in parallel
• Because there is no filename parameter, Codex often ends up:
  • Giving both images random filename
  • Both images saved in same folder
  • One agent ends up referencing/using the wrong image generated by the other agent

This creates race conditions and broken workflows in any project that uses multiple agents or parallel image generation (very common in larger codebases, design systems, game asset pipelines, etc.).

With the requested feature this would be solved instantly:

image_gen(
  prompt = "...",
  size = "1024x576",
  filename = "assets/hero-banner.png"     # ← explicit per-agent control
)

[chrm52]

agent in codex cli going through all the images to find the image which it requested to be generated via imagen. too much of wasted resources. easily fixed if imagen accepts prompt and filename for the generated image or let the agent know the filename of the generated image if its random.