Gemini CLI for social media carousels: setup steps and a speed we have not measured

Can Gemini CLI write social media carousel text?

Yes, the app supports it. Gemini CLI is Google's command-line tool. In Fyslide you can pick it, next to Claude Code CLI, Codex CLI, Ollama, an Anthropic API key and an OpenAI API key, and the app asks it for the slide texts. The app charges nothing for AI, and Gemini CLI uses your existing subscription or account.

What we cannot tell you is how fast or how good it is. We timed Ollama, Claude Code CLI and Codex CLI on 4 November 2026. We did not time Gemini CLI, and we rated no provider's text. So there is no Gemini number to quote here.

What do we know and what don't we?

How do you set up Gemini CLI as the writer?

1. Install Gemini CLI and sign in by following Google's own instructions. This page does not repeat them because they change.
2. Prove it works outside the app. Open a terminal, send it one short prompt and read the reply.
3. Select Gemini CLI as the AI. The Gemini CLI guide and the AI setup docs show where.
4. Create one slideshow from a built-in design on a topic you know, and read every slide.
5. If the app cannot reach the CLI, start with the guide above and go back to step 2.
6. Only after one clean slideshow, run a bigger job.

To let Gemini CLI drive the app through MCP instead of acting as the writer, note that Gemini CLI is not covered by the app's one-click MCP Connect. The MCP docs show the JSON entry to add to ~/.gemini/settings.json.

How can you measure its speed yourself?

Use the same job we used for the other AIs, so your results are comparable with each other. The job from the speed page: topic "morning routines that actually stick", the built-in Bold outline design, six slides per post, call to action on, English, and pictures from a local collection with no image downloads.

Run it for three slideshows, twice. Close heavy programs first. Write each time in the log below.

Date: {date}
AI: Gemini CLI {version}
Job: 3 slideshows of 6 slides, topic "morning routines that actually stick", Bold outline design
Run 1 seconds: {seconds}
Run 2 seconds: {seconds}
Failed slideshows: {number}
Other programs running: {list}

Our own three-slideshow runs came from one Apple M5 Pro Mac with 48 GB, with other agents running:

Do not compare your Gemini time with our numbers. Your computer, your network and your account differ. If you want a comparison, run the same job with a second AI on your own machine and compare your two results. The full test notes are on the speed page, and Codex CLI for slideshows shows when the slowest measured option still fits a workflow.

How do you read your own results?

Expect some spread. Our two runs per AI were never identical, so one run proves little. If your two runs differ a lot, run a third and note the slowest. Compare Gemini CLI only with other AIs you ran on the same computer, in the same session, with the same job.

Then decide what the numbers mean for you. A slow writer is fine for a batch you start and leave, and a poor fit for retrying one hook again and again. The same logic applies to every option in the app.

What should you judge besides speed?

• Length. Does each slide fit your word budget? See words per slide.
• Voice. Does it sound like you or like a template? Making AI slide text sound human lists what to fix.
• Repetition. In a ten-slideshow job, do the hooks repeat each other?
• Failures. Did every slideshow finish, and did any error mention limits or sign-in?
• Screenshots. If you build designs from screenshots, does the design come out usable?
• Account limits. Did a plan or quota limit stop a big job? We have not checked Google's limits for any plan.

What does it cost?

The app does not charge for AI. Gemini CLI runs on your own account, so the cost is whatever that account or plan involves. Bring your own AI key explains how the six options bill differently, and it shows how to measure cost per job instead of guessing.