Mind Journeys 1.8, Chasing Search Traffic, and Teaching My Video Tool an AI Twin

Last week was a release week, mostly content. This week split in two. Mind Journeys got a new player and then a run of small growth changes, each one backed by a number I didn't like. Then, over the weekend, I went back to my video tool and gave it a motion studio and an AI version of me. language-learn got one hard lesson about speakerphones.

Mind Journeys 1.8: A Player Worth Looking At

The old player was a glowing orb and a timer. 1.8.0 replaces it with what I've been calling the halo: the journey's photo full-bleed behind everything, a frosted disc with the time left and the current band and beat, and a progress ring that breathes on the old orb's slow 4.8-second rhythm. It never pulses at the beat rate, and it holds still under Reduce Motion.

The part I'm proudest of is the curve underneath. BandAxisCurveView draws the whole journey on a log frequency axis, with the brainwave bands as zones, the band names down the side, and the phase names along the top. You can see where you are, where you're heading, and drag to scrub. Phases now have names too, so the subtitle reads "Descend · phase 2 of 3 · Delta in 5 min" instead of a bare number.

1.8.1 followed a day later with a Spotify-style player sheet and journey cards with a frosted top and foot, the curve drawn in the band's own colour. The site's screenshots got refreshed to match.

Growth Work, One Uncomfortable Number at a Time

Then came small changes, each one started by a stat that annoyed me.

Zero ratings in every storefront. The app asked for a review after two finished journeys. In the last 30 days, five real users finished even one. So the prompt basically never fired. Now one finished journey earns the ask, and so does saving a new journey. Saves happen behind a builder or a paywall, so the ask waits and Home makes it the next time it settles.

Every purchase came from onboarding. After the first paywall, almost nobody saw another one. So Home now shows a "Shape your own journey" card to people who aren't subscribed. Building and listening stay free; saving is what Pro unlocks. I've also started marking the ten 1.7 journeys as premium behind their own placement. Because the gate fails open, whether they're actually paid is a dashboard switch, not a code change. I like decisions I can undo without a build.

The searches that converted. Apple Ads showed that "lucid dreaming" and "out of body experience" searches turned into installs. So there's now a Custom Product Page aimed at exactly those people, with four screenshots and headlines in their words. It's created in App Store Connect and waiting for submission.

The Site: Answer the Query, Then Get Out of the Way

Search Console gave me the most humbling number of the week. "binaural beats generator" is the site's single biggest query, and the generator page ranked 87th for it. The actual tool sat under two paragraphs of prose. Now the tool is the first thing under the heading, and on a phone every control is above the fold. I measured it in the browser instead of guessing.

I did the same thing to the brainwave guides. Theta got rebuilt around what people actually search for: the 4–8 Hz range first, a band table, what it looks like on an EEG, and an honest section on the evidence. Then I wrote new guides for alpha, beta and gamma (40 Hz), so all five bands now link to each other.

On YouTube, seven new videos went live with journey-card thumbnails. External links are the channel's top traffic source, yet the site got zero clicks from it in 28 days, because every link sat below the fold. So all 12 videos have draft descriptions with the app and the matching guide up top. I haven't applied them yet.

video-tool: Motion, Style Kits, and an AI Twin

Over the weekend I returned to video-tool. Claude now writes an editable storyboard, Remotion renders it locally, and the result drops into the main edit. Style kits (Classic, Archival editorial, Paper collage, Bold tech, or your own) set fonts, colour, motion and captions for a whole cut, and a Claude "director" pass can place titles, lower thirds and cutouts. Missing images come from Gemini or Higgsfield, keyed locally, with a per-project allowance so I don't wake up to a bill.

Then the strange part. With ElevenLabs for voice and HeyGen for a lip-synced presenter, "Remake with my twin" turns any cut into a linked AI-twin project, voiced from the words I kept. The decision I care about most is the badge. Every run of clips is marked REC for real footage or AI for the twin, and exports record where they came from. If I'm going to use an AI version of myself, viewers should always be able to tell.

language-learn: The Tutor Was Hearing Itself

Live calls had two problems. First, the speech-to-text turned accented Japanese into English look-alikes, like "Hi." for はい, and the tutor corrected learners for words they never said. Now it reads transcripts generously and asks about the part it missed. Free practice calls also get end-of-call feedback now.

The second problem took a wrong turn. I built a "speak English" button that swapped speech-to-text streams mid-call, and it worked. Then a real call on speakerphone showed me the actual bug: the tutor's own voice was leaking back into the mic and being transcribed as the learner. I reverted the button and wrote an EchoGuard that drops any transcript that matches what the tutor just said and started while it was talking. Fixing the right problem beat shipping a clever one.

Closing

Most of this week came from looking at a number, disliking it, and making one small change aimed straight at it. Next week I find out which numbers moved.