Measurements
How fast Fyslide makes ten slideshows, per AI provider
Ten slideshows from one topic took 88 and 102 seconds with a local Ollama model on a Mac. Claude Code took 3.5 to 4.5 minutes and Codex about 7 minutes for the same job.
The short answer
- Ollama (local): ten slideshows in about 1.5 to 1.7 minutes (88 and 102 seconds).
- Claude Code CLI: ten slideshows in about 3.5 to 4.5 minutes.
- Codex CLI: ten slideshows in about 7 to 7.5 minutes.
Every post of every run finished; none failed. That is where the home page line “10 slideshows in 2 minutes” comes from: it is true with a local Ollama model on this Mac, and other AI providers take longer.
Raw numbers
Wall clock from the Create request to the job’s final status. Per slideshow = wall clock divided by finished posts.
| Provider | Slideshows | Run | Wall clock | Per slideshow | Finished / failed |
|---|---|---|---|---|---|
| Ollama (local) | 3 | 1 | 36.7 s | 12.2 s | 3 / 0 |
| Ollama (local) | 10 | 1 | 88.0 s | 8.8 s | 10 / 0 |
| Ollama (local) | 10 | 2 | 101.6 s | 10.2 s | 10 / 0 |
| Claude Code CLI | 3 | 1 | 75.6 s | 25.2 s | 3 / 0 |
| Claude Code CLI | 10 | 1 | 4 min 19 s | 25.9 s | 10 / 0 |
| Claude Code CLI | 10 | 2 | 3 min 27 s | 20.7 s | 10 / 0 |
| Codex CLI | 3 | 1 | 2 min 39 s | 52.9 s | 3 / 0 |
| Codex CLI | 10 | 1 | 7 min 14 s | 43.4 s | 10 / 0 |
| Codex CLI | 10 | 2 | 7 min 24 s | 44.4 s | 10 / 0 |
Setup
- Measured on 2026-11-04 with the packaged Mac app, Fyslide 1.4.0 (same generation code as the release, built for testing).
- Machine: Apple M5 Pro, 48 GB RAM, macOS. Other agents were using the machine (and the Codex backend) at the same time, so treat the numbers as typical, not best case.
- Job: Create, topic "morning routines that actually stick", built-in design Bold outline, 6 slides per post, CTA on, English.
- Images: a local collection of 16 sample pictures (no network, no image API, no Pinterest, no AI images).
- Ollama (local) 0.33.3: llama3.1 (8B, 4.9 GB), the app's default text model, running on the same Mac.
- Claude Code CLI 2.1.317: CLI default (no model flag).
- Codex CLI 0.166.0: gpt-6-astra, reasoning medium (from the user's Codex config).
- Gemini CLI was not installed on the test machine and was not measured. The Anthropic and OpenAI API keys were not measured either.
Method
- Start the app with a fresh, empty data folder and activate it.
- Create a collection, import the 16 sample pictures, pick the first built-in design.
- Send one Create job (topic, count, picture collection, provider chosen per job) and poll its status every 0.5 seconds until it is done.
- Plan per provider: three slideshows once, ten slideshows twice, back to back in one app session.
Limits
- Posts are written one after another, so time grows roughly linearly with the number of slideshows.
- Results depend on your computer, the model and the provider’s load. Ollama speed depends on the Mac; the model was already loaded for the second run.
- Only one topic, one design and six slides per post. More slides, web images (Pinterest) or AI-generated images take longer.
- Two runs per provider is a small sample: read the numbers as typical, not as a guarantee.
Try it yourself
Fyslide is a free download; it asks for a license when it starts. Set up Ollama with the Ollama guide, or another provider in the AI setup docs, then create ten slideshows from one topic and time it on your own computer. Download Fyslide or see the pricing.