# How I recorded my first product demo in 54 seconds (OBS, ffmpeg, openai.fm)

> Source: <https://dev.to/serhii_kalyna_730b636889c/how-i-recorded-my-first-product-demo-in-54-seconds-obs-ffmpeg-openaifm-50eg>
> Published: 2026-08-23 04:59:37+00:00

I shipped my first product demo video this week for Convertify (my free image converter). 54 seconds, no team, no budget. Here is the exact toolchain and what I would change next time.

**OBS Studio** for screen recording at 1920x1080, 60fps, CRF 20.

The 60fps is not vanity. Product demos live and die on smooth scroll and cursor motion. 30fps for a converter tool with drag-and-drop looks laggy even when the tool is fast. CRF 20 instead of the default 23 because text on white backgrounds compresses ugly at higher CRF: you get that faint mosaic around letter edges.

Free, open source, deterministic. No mobile screen recording because I needed the desktop browser context (this is a "your iPhone photos on Windows" story).

**ffmpeg** for trim and crop.

```
ffmpeg -i raw.mkv -ss 3.2 -to 57.4 \
  -filter:v "crop=1920:1000:0:40" \
  -c:v libx264 -crf 20 -c:a copy trimmed.mp4
```

CLI over Kdenlive or DaVinci Resolve because I knew the exact cuts. Two ffmpeg commands beat 20 minutes of GUI timeline scrubbing. Crop removed the taskbar and top window chrome without re-encoding twice.

**openai.fm** for voiceover. Ash voice.

I tested Alloy, Fable, Nova, Ash. Ash landed clearest for technical explanation pacing. Alloy sounded like it was reading a bedtime story, Fable had a slight British lilt that felt wrong for the American tech audience I am targeting, Nova was too energetic.

Hiring a voiceover artist starts at $50 for a one-off and rises fast if you iterate. openai.fm generates in 20 seconds per attempt. I regenerated the voiceover 6 times across script rewrites and it cost nothing.

**ffmpeg merge** as the last step.

```
ffmpeg -i trimmed.mp4 -i voiceover.mp3 \
  -c:v copy -c:a aac -shortest final.mp4
```

Separate audio and video passes. If voiceover timing slips, re-recording audio alone is 30 seconds. Re-recording the screen is 10 minutes plus 3 blown takes.

v1 was 180 words. It opened with "Convertify is a free tool that converts HEIC files to JPG." I wrote it, recorded a test take, listened back, and it was plodding. It sounded like I was describing a product, not solving a problem.

v4 is 118 words. It opens with "Your iPhone saves photos as HEIC format. Windows does not. Neither do most websites, Slack, or Discord."

The pattern that unstuck me: **start with the user's pain, not the product name**. The product name earns the right to appear once the pain is on screen. In 54 seconds you cannot afford to spend 5 seconds on brand introduction before the viewer knows why they should care.

Four rewrites to arrive at something obvious. That is the honest part.

**Script cursor movements ahead of time.** I improvised on takes 1 through 4 and each felt slightly different, which made editing harder. Take 5 with a mental map was noticeably tighter.

**Test all voices first, not sequentially.** I burned an hour on Alloy before realizing I should have generated one line in each voice on day one.

**Record the screen segment before writing the voiceover, not after.** I wrote v1 first, then discovered the cursor motion was faster than the words. Rewriting to fit visible pacing was easier than the reverse.

Convertify is at [convertifyapp.net](https://convertifyapp.net). Free, no signup, no watermarks, batch up to 20 files.

Feedback on the video welcome, especially if you have shipped product demos as a solo dev and hit different tradeoffs.