Ollama carousel generator: write slideshow text locally, with measured speed

Can Ollama generate carousels on your own computer?

It can write the text. Ollama runs AI models on your computer, and Fyslide can use it as the AI that writes slide texts. Nothing about that step needs a cloud service or an API key when you use a model downloaded to your computer, and Ollama itself is free. The images come from your own library, the app runs locally, and the slides are rendered with Google Chrome on your machine.

This is the lowest-cost route: the app charges nothing for AI, and with Ollama nobody else bills you either. The price you pay is speed and quality that depend on your hardware and the model, which is why the next section shows numbers.

How fast is local slide writing?

These are Fyslide's own measurements from 4 November 2026 (app 1.4.0, Apple M5 Pro with 48 GB). Each run created ten slideshows of six slides from one topic.

No slideshow failed in any run.

Read the table with its limits in mind.

• Typical, not best case. Other agents were using the machine at the same time.
• No image time. Pictures came from a local collection of 16 samples with no network, no image API, no Pinterest and no AI images. Those would add time.
• One job. The topic was "morning routines that actually stick", with the built-in Bold outline design, six slides per post, a call to action and English.
• Speed is not quality. We timed the runs and did not rate the text. A faster provider is not a better writer.

The full method is on the speed page.

Which model and which computer?

We measured one combination: llama3.1 8B (the app's default text model, a 4.9 GB model) on an Apple M5 Pro Mac with 48 GB of memory. Nothing else is measured. That includes other Macs, other models and Windows, whose version is a Beta that has not yet been tested on real Windows hardware.

What changes speed, per the test notes: the computer, the model, the provider's load and the slide count. Common expectations, not Fyslide measurements:

• A smaller model needs less memory and usually answers faster.
• A larger model usually follows instructions better but needs more memory, and it slows down or fails when memory runs out.
• The first request after starting Ollama can be slower, because the model has to load.

Writing text does not need a vision model. Turning screenshots of a format into a design does. For that you need a vision model, for example qwen2.5vl in Ollama, because text-only models cannot read screenshots.

How do you set it up?

1. Install Ollama and pull a model. The Ollama guide has the commands and what to do when it says "not running".
2. Select Ollama as the AI in the app. The AI setup docs show where.
3. Create one slideshow on a topic you know well, from a design you picked.
4. Read every slide. Check the length against your word limit (see words per slide) and look for repeated phrasing.
5. Only then run ten from one topic.
6. Export PNG or JPG slides, the caption and the zip, and upload them yourself.

Local text still needs an editor's eye. Making AI slide text sound human lists what to fix.

What does local cover, and what does it not?

The app works offline after activation. With Ollama and your own pictures, a whole slideshow can be made without a connection.

When is Ollama the wrong choice?

• You want the strongest wording available and are happy to use a paid or cloud model. Claude Code, Codex, Gemini CLI and API keys are the other options, and each is billed by its provider or your plan.
• Your computer has little memory. A model that does not fit will crawl or fail.
• You need to read screenshots and only have a text-only model installed.
• You are on Windows and need predictable behavior. The Beta has not been tested on real hardware.

Are other slideshow tools local or bring-your-own-AI?

"Local AI" and "bring your own key" are different things.

• carousel.bot (opens in a new tab) describes a browser editor that works locally, with optional AI editing through an MCP companion. It offers export only, and we could not find a pricing page.
• aiCarousels (opens in a new tab) Pro Lifetime runs on your own OpenAI key, which stays in the browser. The AI itself still runs at OpenAI.
• ReelFarm (opens in a new tab) lists bring-your-own AI only on its $195 tier, as an Anthropic API key.

For a lifetime-price view of these choices, see lifetime license versus subscription. For driving the app from an agent instead, see Claude Code and TikTok slideshows.