Codex App built-in image_gen lacks deterministic gpt-image-2 output dimensions for generate and edit workflows

What variant of Codex are you using?

Codex App on macOS (26.417.41555 (1858)), Pro

What feature would you like to see?

I need the built-in image_gen / image_generation path in Codex App to support explicit, deterministic output dimensions for GPT Image workflows.

In ChatGPT web I can generate high-quality GPT-image-2 images and edit/iterate them while choosing specific output dimensions. In macOS Codex App, that does not appear to be available in the built-in path. In practice, this makes my workflow impossible, because I need exact dimensions on both the initial generation and subsequent edits.

From the public Codex repo, the current built-in tool surface still appears to expose only:

{ "type": "image_generation", "output_format": "png" }

Relevant file:

• codex-rs/tools/src/tool_spec.rs

By contrast, the CLI fallback now defaults to gpt-image-2 and already supports arbitrary valid WIDTHxHEIGHT sizes, with validation rules for:

• max edge <= 3840
• both dimensions multiples of 16
• aspect ratio <= 3:1
• total pixels between 655,360 and 8,294,400

Relevant file:

• codex-rs/skills/src/assets/samples/imagegen/scripts/image_gen.py

What I need from the built-in Codex App path:

• explicit size support on built-in image_gen
• arbitrary valid WIDTHxHEIGHT, not only a few presets
• exact size control on both initial generation and edit iterations
• deterministic output sizing instead of prompt-only guessing
• clear validation errors for invalid sizes rather than silent fallback

This is the parity gap I’m trying to close:

• ChatGPT web: exact-dimension GPT Image workflow available
• Codex CLI fallback: gpt-image-2 size control available
• Codex App built-in image_gen: no explicit dimension control exposed

Additional information

Related issues:

• #18636 Ability to reliably generate transparent images
• #18905 Codex image generation saves transparent-background requests as opaque RGB PNGs
• #18944 Built-in image_gen is missing basic image controls like transparent background, size, quality, and masks (closed as duplicate)

Those threads are related, but my blocker is specifically deterministic built-in dimension control in the macOS app workflow, for both generate and edit.