FAQ

How do I use gpt-image-2 in Codex?

You do not need to switch the Codex model to gpt-image-2.

Do not use gpt-image-2 as the Codex main model. It only generates images and cannot replace a normal conversation or coding model.

Keep gpt-5.6, or another conversation and coding model, as the Codex main model. When you ask for an image, Codex calls gpt-image-2 in the background.

Step 1: Configure Codex normally

Follow the Codex setup guide to connect Codex to ruoli:

Setting	Value
Main model	`gpt-5.6` or another conversation/coding model
API URL	`https://ruoli.dev/v1`
API Key	Your token from the ruoli console

Step 2: Enable image generation

Open CC Switch and go to Codex → Edit Provider → Config JSON.Find the active [model_providers.your-provider-name] section and add these two lines:

requires_openai_auth = false
http_headers = { "x-openai-actor-authorization" = "local-image-extension" }

The highlighted area in this screenshot shows where to add them:

Enable gpt-image-2 in the Codex provider configuration

Save the configuration and restart Codex.

Step 3: Describe the image to Codex

You do not need an API command or a model switch. Ask Codex directly:

Generate a cyberpunk view of Shanghai at night in a 16:9 aspect ratio.

Codex understands the request and calls gpt-image-2 automatically.

Your token must use the gpt-pro group. If you get a “no available channel” error, check that the token uses the correct group.

How do I call the image generation API?

To generate images from your own program, script, or website, call the image API directly. This method does not use Codex.Prepare these three values:

Item	Value
API endpoint	`https://ruoli.dev/v1/images/generations`
API Key	Your token from the ruoli console
Model	`gpt-image-2`

Direct API example

Replace sk-YOUR_KEY with your token, then run:

curl https://ruoli.dev/v1/images/generations \
  -H "Authorization: Bearer sk-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "An orange cat wearing an astronaut helmet, sitting on the Moon, cinematic lighting"
  }'

The text after prompt describes the image. Replace it with your own description.

New to coding? Let an Agent configure it

Paste the following prompt into Codex or another coding Agent. Replace sk-YOUR_KEY with your token:

Create a Skill that calls an image generation model, and test whether it can generate an image successfully.

API endpoint: https://ruoli.dev/v1/images/generations
API Key: sk-YOUR_KEY
Model: gpt-image-2

Use the OpenAI-compatible Images API and pass the key with Authorization: Bearer.
Do not write the API Key into the Skill or commit it to Git. Store it in the RUOLI_API_KEY environment variable.
The Skill should accept my image description, call the API, and save the generated image.
When finished, explain in plain language how I can use the Skill again.

Never send a real API Key to strangers or commit it to a public repository.

How do I enable Fast mode in Codex?

Think of Fast mode as a priority lane for Codex. When enabled, requests use the priority service tier.

Open the Codex Config JSON

Open CC Switch and go to Codex → Edit Provider → Config JSON.

Add the Fast mode setting

Find the general settings near the top, alongside options such as model and model_reasoning_effort, then add this line:

service_tier = "priority"

The highlighted area in this screenshot shows where to add it:

Set service_tier to priority in the Codex Config JSON

Put service_tier in the top-level general settings. Do not put it inside [model_providers] or a provider section.

Save and restart Codex

Click Save, fully quit Codex, and open it again. Fast mode will take effect after the restart.

To turn Fast mode off later, remove the service_tier = "priority" line and restart Codex.

What should I do if I see Stream disconnected before completion?

This error means the connection ended before Codex finished its response. It does not always mean the relay service is down. Check these four causes in order.

Check your network

An unstable connection is the most common cause. If the network briefly drops, the Codex response can stop as well.Try the following:

Turn your proxy or VPN off and on again
Switch to another proxy node
Switch between Wi-Fi and a mobile hotspot
Send the same question again after the connection recovers

If the retry works, the cause was probably a temporary local network issue.

Check whether the conversation is too long

A longer conversation gives Codex more context to send with every request. The request may fail when that content exceeds the model’s context limit.The easiest fix is to start a new conversation and ask again.If the new conversation works, the old one was probably too long. To preserve important information, ask Codex to summarize it first, then paste the summary into the new conversation.

Check whether the request triggered a safety refusal

OpenAI may refuse requests involving illegal actions, malicious attacks, dangerous behavior, or other high-risk content. The connection may end before a complete answer is returned.Check whether your question could be interpreted as harmful. Rewrite it to describe a legitimate, safe purpose. Do not try to bypass the model’s safety restrictions.

Check the relay service

If the network is stable, a new conversation still fails, and the question is safe, the relay channel may be temporarily unavailable.Check the group messages for an outage notice or maintenance announcement. If other users are reporting the same error, wait for the service to recover and try again.

Check the causes in order. The first three are also common, so this error does not automatically mean the relay service is down.

Low cache hit rate or unexpectedly high cost in Claude Code?

It’s almost always the default attribution header — every request carries a dynamic session ID that blows out the upstream prompt cache.In CC Switch → Edit Provider → Config JSON, add one line to env:

"CLAUDE_CODE_ATTRIBUTION_HEADER": "0"

Recommended whether you’re on native Claude or a third-party model. See Disable Attribution Header for details.

Quick Start

Platform

Terms of Service

Step 1: Configure Codex normally

Step 2: Enable image generation

Step 3: Describe the image to Codex

Direct API example

New to coding? Let an Agent configure it

​Step 1: Configure Codex normally

​Step 2: Enable image generation

​Step 3: Describe the image to Codex

​Direct API example

​New to coding? Let an Agent configure it

Step 1: Configure Codex normally

Step 2: Enable image generation

Step 3: Describe the image to Codex

Direct API example

New to coding? Let an Agent configure it