Kenneth W. Bingham

portfolio project · JavaScript · Electron · Web Audio · MediaPipe · local AI

A podcast studio where the episode is finished when you stop talking.

ButterflyFx Podcast Creator lets you write the visuals into the script. Every graphic, caption, transition and effect appears on the exact word you say, while a teleprompter keeps that word right under the camera so you are always looking at the lens. No post-production.

Watch it being used

A 69-second walkthrough of the real studio: open an episode folder, see the cues in the script, set the look, then talk. The marquee teleprompter keeps the current word under the camera, and each graphic and caption appears on the word that calls for it. Narrated by an Australian neural voice; recorded straight from the studio's screen.

The episode it made

A 75-second episode recorded by the studio itself, with no person on camera and no editing afterwards. The host is an illustration, and every picture and caption was placed by the script.

The problem

Most video podcasts are recorded first and dressed afterwards. Someone scrubs through the footage, finds the moment the host mentions the chart, drops the chart in, and does that again for every caption and transition. It takes hours, and the pictures rarely land on the word that needs them. Meanwhile the host reads from notes off to the side, so they look like they are reading.

What it does

Here is part of the script behind the demo above:

A cue names {text ll | Where it goes} where it goes, {cleartext} {text ll | Which asset} which asset to show,
{cleartext} {text ll | A short label} and a short label.
{clear} {news | placement.svg | Cue before the word} The one rule: put the cue right before its word,
and the picture lands the moment you say it.

How the demo was made

The narration is a Microsoft neural voice, which reports the exact moment each word begins. Those moments drove the studio the same way its voice-follow mode tracks a live speaker: each word advanced the teleprompter and fired that word's cues, and each narrated step clicked through the real interface, while the browser recorded the screen. The same steps can turn any script and a folder of pictures into a finished episode, which is how I plan to make demo videos of my other projects.

The proof

Technologies and skills

What it does not claim