Image Generation for Web Projects

[nectario]

Feature Request: Image Generation Support for Web Projects

It would be extremely helpful if codex CLI could leverage GPT-4o’s image generation capabilities to automatically create images for web projects during development.

Details:
When working on a website, developers often need placeholder or final images for elements such as banners, product photos, icons, or hero sections. Instead of manually sourcing or designing these assets, it would streamline development if codex could generate relevant images based on short natural language prompts directly in the CLI or through a codex web assets command.

Proposed Behavior:
• When in a web project context, codex can offer an optional --generate-images flag or similar.
• The CLI can ask the user what type of images they need (e.g. “a modern tech product banner”, “a clean login form illustration”, etc.)
• GPT-4o would then generate and save the images into the project’s assets or images directory.

Why This Matters:
This feature would significantly reduce friction for solo developers, designers, and prototypers by eliminating the need to switch tools or hunt for royalty-free images. It aligns with codex’s goal of speeding up the development lifecycle using natural language interfaces.

[potatoqualitee]

Yes please 🙏🏼

[amitbcp]

Yes please

[kquinsland]

I am a +1 for this. Usecase is akin to "take a look at this_script.py and identify its core function. I need a 512x512 image that best represents this core function, theme/style should be consistent with @path/to/other/style-guide.md"

[darrenoakey]

please please add this - any app that you write nowadays - be it a web app or an iPhone app or whatever - or even just making the icon for a completely backend thing - needs some sort of visual assets. It's crazy that we can do everything else - setup, deploy, actually write the code, fill out the iras etc from the cli - but not actually generate the images. well.. I do actually - I have a cli generate_image which uses mflux and qwen or z-image to generate images, and the cli calls that - but I would much rather have the quality and speed that comes from chatgpt-image

[etraut-openai]

This is a discussion forum, so it's a good place to ask questions or share usage tips. If you want to suggest a new feature, it's better to file an enhancement request in the issue tracker. The codex team uses upvotes on enhancement requests to inform feature prioritization.

[amitbcp]

@etraut-openai can you please help in doing that ? really want to add this to feature request

[etraut-openai]

Anyone can file an enhancement request. Simply go to https://github.com/openai/codex/issues and click the "New Issue" button. Then fill in the template for an enhancement request.

[MKcodeshere]

Yes it is really helpful for lots of use cases - bring it on

[lSelectral]

Please this is very needed

[kquinsland]

Codex will happily generate svg filess. It will also give you a simple svg to X converter python script. Not quite as good as arbitrarily images from nano banana or similar but it can work in a pinch.

[nectario]

One workaround I use is asking Codex to generate images using an API call. I've been using it so often like this that I forgot this issue even existed.

[nectario]

Workaround (today): call the latest OpenAI image model directly via API

Codex doesn’t currently wire up the newest image generator in the CLI flow, but you can still use the latest image model by calling the OpenAI Images API yourself (and then saving the result into your project’s assets/ folder).

Model: gpt-image-1.5
Endpoint: POST /v1/images/generations (or /v1/images/edits for transforms)

1. Text → Image (generate)

export OPENAI_API_KEY="YOUR_KEY"

curl https://api.openai.com/v1/images/generations
-H "Authorization: Bearer $OPENAI_API_KEY"
-H "Content-Type: application/json"
-d '{
"model": "gpt-image-1.5",
"prompt": "A clean modern hero banner for a fintech dashboard website, minimal, flat illustration, lots of whitespace",
"size": "1024x1024"
}'

The response includes base64 image data — decode it and write it to something like:
./public/assets/hero.png

2. Image → Image (edit / transform)

curl https://api.openai.com/v1/images/edits
-H "Authorization: Bearer $OPENAI_API_KEY"
-F "model=gpt-image-1.5"
-F "prompt=Make the background transparent and keep the style flat and minimal."
-F "image=@input.png"

Quick notes
• This is the same “latest” image stack available in the API even if Codex isn’t using it yet.

Docs (for reference):

https://developers.openai.com/api/docs/guides/image-generation/
https://developers.openai.com/api/docs/models/gpt-image-1.5

[nectario]

But honestly, if you tell codex to use the latest image API it knows how to do it.

[darren-starrez]

there are lots of apis - clearly its trivial to just use an api key - and you might as well use nano banana if you're going to do that. The entire point is to avoid cost risk by doing a generation within the current subscription.

[shushenghong]

if i have subscribed a codex plus / pro plan, do i need pay when use api key to generate images?