See the work before you pay.
Everything here is something we made and put out in public: our own daily show and its Short, a before and after audio sample, narration demos, and a tool we built in the open. Each one comes with its file specs and the real numbers.
Free your own script, narrated · up to 150 words · back within 24 hours
Nothing to pay.
01Podcast package
One episode in. Every piece out.
Prompt Daily is our own daily tech news show. Episode 17 came out on Saturday, October 17, 2026. From one finished audio master we made the YouTube video, a caption file, chapters and a vertical Short, and published all of it.
Full episode, 1920×1080, with an English caption track and 9 chapters on YouTube.
The show is AI-voiced. The hosts, Maya and Ravi, are synthetic voices, and every episode says so. The full video is how we publish our own show. For your episode, the $29 Launch Pack covers the master, full-episode captions and 3 vertical clips.
The Short: 1080×1920, captions timed word by word, cut straight from the master.
- Video length matches the audio master: 881.43 s vs 881.42 s
- Audio in sync with the master at 6 points: 0 ms off
- Caption file: 283 cues, no overlaps, no zero-length cues
- Longest caption line: 42 characters
- Short: 48.6 s clip plus a 2.6 s end card, under 58 s
- Short audio matches its slice of the master: yes
- Chapters: 9, all at least 10 s apart
- These are structure numbers. Cue timing and line length don't show that the captions are word-for-word accurate.
Selected rows from the episode's real file log, Oct 17, 2026.
Captions
283
cues in the caption file we made for the episode. Lines stay at 42 characters or fewer.
Chapters
- 0:00Cold open
- 0:43OpenAI's agent took a wallet
- 5:05Tesla's flying Roadster
- 6:47ChatGPT drives your Chrome tabs
- 8:09Rocket Lab Neutron slips to 2027
- 9:34Week in review
- +3 more
Short
0:51
"Neutron slips again," cut from 8:09 to 8:58 of the master, with an end card that points back to the full episode.
02Before and after
One clip, before and after.
Press play, then flip between the two. It keeps your place, so you hear the same words both ways. This is the loudness step from Podcast Polish: a quiet file brought up to podcast level with the peaks kept in check.
Before
−22.5
LUFS · peak −3.1 dBTP
After
−16.0
LUFS · peak −1.6 dBTP · mono
What this sample is and isn't: it's the first 30 seconds of our own narration demo, which was published at a quiet −22.5 LUFS. It shows loudness only. It isn't noise repair or multitrack mixing, and Podcast Polish doesn't do those. Both files were measured with ffmpeg's EBU R128 meter.
Audio only in, vertical clip out
No video? You still get clips.
This 38-second clip started as an audio file with no picture. It got a clean background, a live level meter and captions timed word by word, with the spoken word lit up. Delivered clips are full 1080×1920 MP4s. This one is a smaller preview so it loads fast.
Refund policy: one revision round per episode; cancel by email, current month not refundedRefund policy: first episode free, one revision on every order03Narration
Hear the voices on real scripts.
Two of our demos, rendered start to finish through the same steps as a paid order. The explainer is also where our sample caption file came from: 21 cues, no line over 42 characters, no cue faster than 17.2 characters per second.
True crime · 76 s
Documentary cold open
An original cold case, told tense and cinematic. Narrated by Charon.
Technical explainer · 82 s
Semiconductor deep dive
Dense facts, read clearly: why extreme ultraviolet lithography was so hard. Narrated by Kore.
Every narration order is read from your exact script. Flag any word we misread and you get one free re-render pass. Full narration is $29 per finished hour (about 8,700 words), or from $19/month. All four demos
04Automation
We build our own tools. Here's one, in public.
Taint + Receipts is a tool we built for the Graph Hacks hackathon (WeMakeDevs and FalkorDB). Say one wrong number (3.5 hours read as 35) spreads into a blog post, a landing page, an email and a Short. Fixing the number takes a second. Finding every place it went is the hard part. This tool marks everything built on the bad number, sorts it by how many people likely saw it, explains why each piece is wrong down to the source line, and queues the fixes.
Demo video, 1:51. It tracks a made-up product, Acme Notes. All names, claims, prices and numbers in the demo are made up.
10
pieces in the demo traced back to one wrong number
10 hops
deep, the furthest a correction follows the trail
0.55 ms
median read-only downstream count, up to 10 hops, on a synthetic 26,004-node, 51,099-edge test graph (0.62 ms on a later run)
~14.9 ms
local correction query plus recall plan on that graph, not counting receipt checks or model and API calls
Timing numbers come from one shared Linux box, so yours will differ. The test script is in the repo.
Straight answers
What this page doesn't show.
Files, not quotes
We only post what we can back up. So this page sticks to files you can play and numbers you can check.
Our show, not yours
Prompt Daily is our own production. Your episode gets the scope on the page you order from, nothing more and nothing less.
The best test is your script
Demos only go so far. The free sample narrates up to 150 words of your own script, so you judge it on your words, not ours.
05 · Try it free
Hear it on your own script.
Send up to 150 words of your script and pick a voice. Within 24 hours you get it back, narrated and mastered.
- Up to 150 words of your script
- Back within 24 hours
- Narrated and mastered
- Nothing to pay
Then $29 per finished hour, or from $19/month. Podcasters: the $29 Launch Pack covers one episode up to 60 minutes.