JFK on the decks
A president behind a club mixer, in ten seconds.
Phi Pham, creative technologist. Sora and Veo generation, word-timed kinetic type, cutout collage and code-drawn animation, cut by an editing pipeline I built. No CapCut, no templates. One clip passed 1.1 million views.
Muted previews. Full pieces are linked in the notes below. Scroll the strip sideways.
An experiment account: politicians and moguls doing things they never did. Ten-second clips, generated in Sora and Veo, written and cut for the feed.
The account has 1,357 followers. The clip reached 1.1 million, so the reach came from the feed, not from an audience. @boilertalkn ↗ @kawaiithoughts0 ↗
What's in each piece, what made it, and how it did.
A president behind a club mixer, in ten seconds.
Reagan's most famous line, as a breakdance. Same account, same ten-second format.
A fully generated mukbang channel: eight-second clips built around the "67" meme, on-camera hosts who don't exist.
Talking head plus six generated b-roll shots. One style-anchor image attached to every prompt, so all six read as one film.
Every spoken word lights up as it lands. Numbers count up and archive clips cut in on the line.
Cut-paper collage with me as a sticker in the corner, cut out of every frame of my own footage.
Collage over his own podcast clips, with sound effects placed to picture.
A doodle explainer drawn in code, frame by frame, timed to the narration. From my channel How TMRW Works.
Seven generated elements composited into one mural. An independent AI judge scored every round, and only improvements were kept: 4.2 to 8.4 out of 10 in twelve rounds.
A boombox character drawn and animated in code, with every sound synthesized from scratch. Made for Building Beats.
View counts as of October 20, 2026.
The edited pieces above, 04 to 10, went through the same chain. I built it so one person can ship finished, checked edits on a schedule.
Written for the ear, every claim tied to a source before anything is recorded.
Sora, Veo and Nano Banana, with one style-anchor image on every prompt so shots match.
whisper transcribes the take to the word, and every caption and card hangs off those timestamps.
Cards, captions, cutouts and collage composited in code, not dragged in a timeline.
Voice, music and effects mixed and loudness-matched for the feed.
Automated passes for false claims, captions covering faces, picture-to-sound sync and false starts.
An audio listening pass that reports what plays when, before anyone else hears it.
Vertical masters for TikTok, Reels and Shorts, then the numbers come back into the next script.
Tools: Sora, Veo and Flow, Nano Banana, ElevenLabs, whisper, ffmpeg, Pillow, Apple Vision, OpenCV, canvas and three.js, Claude Code.
A repeatable workflow for your format: generation, editing, captions and QA, documented so your team can run it without me.
Kinetic type, word-timed captions and collage edits for creators and brands who need volume without the template look.
Forward-deployed creative work: sitting with a team, making the thing, and turning what worked into a system.