0 → Claude edits the videos now
The editing loop closes on itself. Snipy, the tool that already pulls a transcript with Whisper, gets wired to CapCut, which turns out to keep its project as a JSON file Claude can edit directly, so Claude makes the cuts and drops in sounds; three videos out in about six hours, and it learns your style as it goes. Juan's tip to Julia is the recursive one this whole blog runs on: once you have three or four edited videos, ask Claude to study them and distill your editing into its own instructions, a glossary of which sound goes where. But the honest counterweight is Julia's: the fun part, for her, is the exploration, hunting a fresh sound on the Myinstants soundboard, and if Claude just reuses the same ten sounds every day the work goes dead. Automating the boring frees the person for the creative, but only if you're careful about which half is which.
// trace: where this idea came from
- ↳ video diary @ 55:57 (conectar Snipy (que saca el transcript con Whisper) a CapCut, que tiene un JSON editable por Claude)
- ↳ video diary @ 57:01 (ya publicamos tres videos, como 5-6 horas, y Claude va atrapando el estilo)
- ↳ video diary @ 1:01:32 (tip, con tres o cuatro videos pídele a Claude que destile tus cortes y sonidos en sus propias instrucciones)
- ↳ video diary @ 59:37 (Julia usa Myinstants, un soundboard viejo de botones; lo divertido es la exploración, no la repetición)
- ↳ Entry 271-2: Gemini called the match (los mismos videos de fútbol cuyos gráficos edita esta tubería)
- ↳ Entry 269-3: The blog you are reading (la misma recursión que rige este blog, ahora aplicada a la edición)
Snipy already did half the job: point it at footage and it pulls the transcript with Whisper. The last idea was to connect it to the rest of the edit, and it turned out CapCut, running locally, keeps its project as a JSON file, a file Claude can open and edit directly ▸ 55:57. So now the loop is: Claude reads the transcript, decides whether it’s one video or two, makes the cuts, drops in sounds, and hands it back. Three videos out already, maybe six hours of work, and, the interesting part, Claude is catching Juan’s style as it goes; he thinks it only gets faster ▸ 57:01.
His tip to Julia is the same recursion this whole blog is built on. Once you have three or four videos edited, don’t just keep editing, ask Claude to study them and extract how you edit into its own instructions: which sound goes where, how the cuts fall, a glossary of your taste written back as a rulebook ▸ 1:01:32. The pattern is the fixed point of the year: do the thing a few times by hand, then have the machine watch and distill the doing into a spec, so next time the machine does more of it.
automatizar lo aburrido, pero cuidado con cuál mitad es cuál →
Then the honest counterweight, and it’s Julia’s, which is why it belongs in the entry. For her the fun part is the exploration: hunting a fresh sound on Myinstants, the old button-board soundboard where anyone can upload new effects, watching for the new button, “what’s this one?” ▸ 59:37. If Claude just glues the same ten sounds onto every video, the work goes dead; she edits three videos a day, and reusing the same ten from here to here for an hour and she’s done, it feels repetitive, she wants new light. To a viewer who sees a clip once it makes no difference; to the editor it’s the whole difference. Which is the real shape of the lesson: automating the boring half frees the person for the creative half, and that only works if you’re honest about which half is which. Juan feels the same pull on his own side, the joy is in reaching the one unique idea, the Pokémon bestiary, the Lego transformer, not in the hundredth repetition. The machine should take the repetition. The exploration stays human…
// continued in
no entry has continued this idea yet: the arc is still open