Built-in image_gen is missing basic image controls like transparent background, size, quality, and masks
What variant of Codex are you using?
Codex CLI / Codex App with the built-in image_gen tool
What feature would you like to see?
What feature would you like to see?
I’d like the built-in image_gen tool to expose the basic image-generation controls that already
exist in the image API / CLI fallback, especially:
background(transparent/opaque/auto)sizequalitynmaskinput_fidelity
Why this matters
The built-in tool is useful, but right now it feels too limited for real asset work.
A few common cases are hard to do reliably without these controls:
- generating a real transparent-background PNG
- choosing the output size / aspect ratio
- doing masked edits
- staying inside the subscription-based Codex flow instead of falling back to API-key-based workflows
What I ran into
I tested the built-in image_gen tool with prompts like:
background: transparenttransparent background
The image looked like it had a transparent-style background, but the saved file was still a normal
RGB PNG with no alpha channel.
So in practice, prompt wording alone does not seem to be enough when you need actual transparency.
Why I think this is missing in the built-in path
From the public Codex repo, the built-in image_generation tool appears to expose only
Relevant files:
codex-rs/tools/src/tool_spec.rscodex-rs/tools/src/tool_registry_plan.rscodex-rs/skills/src/assets/samples/imagegen/references/image-api.md
What I’d expect
If the built-in image tool is meant to be used for real image generation and editing, it should
expose the main controls needed for that work, especially:
- transparency
- output dimensions
- quality
- masks
- edit fidelity
The biggest one for me is background=transparent, because that is hard to replace with prompt
wording.
Additional context
Related issues / discussions:
- #8758
- #14895
- Discussion #592
Additional information
_No response_
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Additional data point from real usage:
For my workflow, the web product and Codex are not behaving the same way.
In the web product, for this kind of asset generation, it is often enough to state in the image-generation instruction/prompt that the output format must be PNG, and the result comes back with usable transparency.
In Codex, that same practical instruction is not enough. Even when the prompt explicitly says things like:
...the saved result still comes back as an opaque PNG or a fake transparency/checkerboard-style result.
That makes this feel like a Codex tool-surface / wrapper gap rather than only a prompting issue.
So I strongly support exposing the actual image controls in the built-in
image_genpath, especially:output_formatbackground=transparentsizequalityIf the underlying image capability can already produce transparency in the web flow, Codex should expose the equivalent control instead of forcing users to rely on prompt wording alone.
This should be the main issue, in my opinion, since it includes all the controls that ChatGPT has for images, instead of just transparency.