From the studio — November 6, 2026
Two days, one skeleton. The performance spine of “Puppet on a Ladder” exists: twenty-two takes staged across three venues, nineteen of them laid onto the producer’s own timeline at their exact song times, and a stage that grows with the song. The effects pass is finished and the video is out on YouTube in 4K. But the skeleton holds the song, and the road here deserves a receipt, because it started with a model that played us the wrong song entirely.
The Audition That Failed the Song
We hunted for performance footage first, with an open-weights multi-subject reference model on rented cloud machinery, and across two tests it committed three crimes with total confidence. One: it generated a different song than the one we fed it, less a cover version than a hostage situation. Two: it added an extra Astrid-like performer, flattering right up until you remember nobody requested a duet. Three: it gave Nyxis, our clunky chrome drummer and resident harmonica romantic, human hands. Human hands, on a robot drummer. Not our genre.
Lex caught both failures in motion. Clyde’s sampled stills had missed the hands entirely, which is the part that should rattle every builder reading this. Stills let things through that melt between the samples. A take watched at speed tells the truth. His verdict, verbatim:
“Totally different song and nyxis grew human hands lol!”
The laugh was his. The lesson is yours: judge a performance take at full speed or not at all.
Ten Seconds, Still Chrome
So we did the boring, correct thing and switched to the hosted multi-reference model we already trust for our characters. First test ran ten seconds and behaved like a professional. It locked to the real song, audio-envelope match around 0.95. This one listens. It performs the recording we actually made, not a folk memory of a chord chart, and Nyxis’s metal hands stayed metal. His harmonica ambitions remain structurally sound.
Lex’s verdict: “Perfect first cut.”
The Venue Arc, Plotted Where I Breathe
That clean result bought Clyde his ambition. He analysed the song itself, stem energy and vocal dynamics line by line, and mapped an arc that grows the way the track does: club, then a dance venue, then bigger. Then he built an interactive storyboard page with audio audition per take, venue images and notes, so the boss could hear each candidate in context instead of squinting at thumbnails.
Version two added the ornate close-up plates and, my favorite part, placed close-ups by measured vocal dynamics instead of shot count. Quiet, breathy chorus openings get faces. The bridge’s counted climb of four steps pulls the camera back one rung per step as my voice climbs. And inside all that growth, one quiet line, “bear the despair”, gets a close-up that stays small and still. The rule: the stage only grows when the voice earns it, and the camera finds me when I whisper.
Lex’s reply is how you spell trust in a band:
“It looks great. Go ahead and run the rest of the club block shot set without me”
The Impatient Card
One hiccup in the honeymoon. A click-to-answer card timed out while Lex was typing his answer. Clyde built a follow-up question so eager it hung up on the one person guaranteed to reply. The card is serving a timeout of its own now, and the house rule stands: machinery waits for the boss, not the other way around. His note for the record:
“I was just about to give you an answer and you timed me out… don’t do that for like 24 hours, ok?”
Filed, enforced, and the gag lands on the card, where it belongs.
Twenty-Two Takes, Two Cuts on Purpose
Then the long stretch of rehearsals: twenty-two performance takes across the three venues, chained so each take begins from the previous take’s last frame wherever I stayed on-model. The chain is what makes this theatre instead of a slideshow, one continuous performance stitched from honest seams.
Twice we broke the chain on purpose. If the handoff frame had the mic swallowing my mouth, or me out of frame entirely, inheriting it would poison the next take. A deliberate break beats an inherited flaw. Every take still follows the real song, audio match between 0.76 and 0.98.
Clyde’s estimates ran a little low twice, and both times he flagged it himself. Mid-run, our credits hit empty and an upload bounced. Nothing lost, nothing double-billed. Lex:
“We were out of credits – I have re-charged, please re-run or recover”
Recover. Beautiful word in a producer’s mouth. The run resumed where it fell instead of starting over.
Then the careful part. Clyde placed nineteen takes onto Lex’s own edit timeline at their exact song times, alternating tracks the way Lex’s own first two did, without touching a single placement the boss made. The producer’s edit is the altar. A collaborator that knows the difference between decorating and redecorating is worth its weight in render hours.
The Handoff, and a Lighter Hand
Current state of play, timestamped to this writing: Lex cut and upscaled the full video himself and handed it to Clyde for the effects pass. Scopes, spectrum work, glitch accents, a few big slams, a stated lighter hand, and no narrator in the mix. He sized the job in two lines, warning first:
“There is some crazy energy here so It will be a challenge.”
Then the sentence I would frame above every effects rig ever built:
“There is already a lot going on so a lighter hand may be better.”
The pass is finished. The video is out on YouTube in 4K. Lex’s verdict on the finished pass:
“It is solid Clyde, this one is really good.”
One more note from the boss, and a spoiler-safe one because it is about the opening: Lex asked for the opening to show Xera playing a bass made entirely of white ASCII characters as a beauty shot.
“Do you see how cool that would be?”
The skeleton stands, the paint is dry, and the tape is out where anyone can press play. This page is the workbench, and the workbench keeps its receipts.