Codex CLI for social media slideshows: measured timing and when slow is fine

Can Codex CLI write social media content for slideshows?

Yes, as the writer behind an app. Codex CLI is OpenAI's command-line tool. In Fyslide you choose it as the AI, and the app asks it for the slide texts. You keep the design, the pictures and the export in the app. Codex CLI is one of six AI options there, and the app charges nothing for AI.

Here is what we measured. On 4 November 2026 (app 1.4.0, Codex CLI 0.166.0, reasoning set to medium, one Apple M5 Pro Mac with 48 GB), two runs of ten slideshows with six slides each, from one topic, took 434.2 and 443.8 seconds. Three slideshows took 158.8 seconds. No slideshow failed. The method is on the speed page.

How does that compare with the other writers?

Every run used the same machine, topic and design. Ten slideshows from one topic, fastest first:

• Ollama, llama3.1 8B: 88.0 and 101.6 seconds.
• Claude Code CLI 2.1.317: 259.2 and 207.4 seconds, about 3.5 to 4.5 minutes.
• Codex CLI 0.166.0: 434.2 and 443.8 seconds, about 7 to 7.5 minutes.
• Gemini CLI: not measured. See Gemini CLI for slideshows.

Read these as typical, not best case. Other agents were using the machine while we measured, and some of them were using the Codex backend. Pictures came from 16 local samples, so no image downloads or AI images are included. Speed is not quality either: we timed the runs and did not rate the text.

When is a slower writer fine?

Ask what you do while it runs.

We timed jobs of three and ten slideshows only. A single slideshow was not timed, so this page does not give a number for one.

What does a weekly routine with a slow writer look like?

Example (a fictional schedule, not a measurement):

1. Pick two topics and start a ten-slideshow job on the first.
2. While it runs, choose pictures for the second topic.
3. Read the first batch, delete the weak slideshows and rewrite the hooks you keep.
4. Start the second job and read while it runs.
5. Export the keepers together and upload them yourself.

In this routine the machine's waiting time overlaps with your own work, so the extra minutes cost you almost nothing. The same job would hurt in a workflow where you sit and watch a progress bar.

What did the test not cover?

• Reasoning settings other than medium. We cannot say what a different setting changes.
• Other Codex CLI versions, other computers and Windows, which is still a Beta.
• Slide counts other than six, and jobs with web or AI-generated images.
• The text quality of any provider.
• Codex CLI driving the app through its local MCP server. The MCP docs describe that route, and we have not timed it.

How do you set up Codex CLI as the writer?

1. Install Codex CLI and sign in by following its own documentation. Check that it runs in a terminal before you open the app.
2. Select Codex CLI as the AI. The Codex CLI guide and the AI setup docs show where.
3. Make one slideshow from a design and read every slide.
4. Time a three-slideshow job on your computer with the log below.
5. Only then run ten, and review them before you export.

In the app, Codex CLI is also one of the vision-capable AIs that can turn screenshots of a format into a design.

Date: {date}
AI: Codex CLI {version}, reasoning setting {setting}
Job: {number} slideshows of {number} slides, topic "{topic}"
Seconds from pressing Create to the last slideshow: {seconds}
Failed slideshows: {number}
Other programs running: {list}

Who pays for Codex CLI here?

Codex CLI uses your own plan or account, so the bill comes from the provider, not from the app. Bring your own AI key explains the three billing models. For a routine that ties several writers together, see batch creating slideshows.