The first 100 members lock 40% off. For life.Claim yours →
Claude · question

Claude Code and Remotion for short videos: what breaks?

The toolchain is free. Keeping it standing is the job.

The short answer

The published workflows agree on where it breaks: renders glitch when several run at once, FFmpeg's drawtext "is unreliable" for on-screen text, a missing libass filter stops captions, a caption library crashes on install, previews show "0:00/0:00", and a first assembly ran "15+ minutes per chapter" until it was rewritten. The first render then "usually needed human correction". RealTake renders in the cloud with nothing to install, and lets you upload the next video while one renders.

What breaks, from the people who built it

  • "Ran into issues midway through a couple of them" rendering five clips at once; renders "take 2 to 10 minutes". (Blotato)
  • "Glitchy face cam because several renders were running"; the localhost preview "shows 0:00/0:00". (Clixie)
  • "Text rendering through FFmpeg's drawtext filter is unreliable"; first assembly "15+ minutes per chapter" until stream-copy brought it to 30 seconds. (Wondering About AI)
  • Homebrew FFmpeg without libass ("No such filter: 'subtitles'"), a faster-whisper and PyAV crash, a libvpx dyld break. (laozhang)
  • Motion design and colour grading rated "Not well". (Sprites)

Why it keeps breaking

Remotion renders React to video frame by frame on your machine; FFmpeg does the cutting; a separate model does the transcription; Claude writes the glue. Four moving parts, each with its own install, versions and limits, and any one of them stops the pipeline. Running two renders at once competes for the same CPU and the output tears. None of this is editing. It is the cost of owning a render farm of one.

How RealTake does it instead

RealTake runs the same job as a product. Upload a raw video of yourself talking; the Director plans the whole edit, cuts the rambles with a reason on every cut, puts word-level captions on the frame, places b-roll and 200+ motion graphics clear of your face, adds sound, and renders in the cloud in minutes. Change anything in the Studio before it renders at no cost, then post or schedule it to LinkedIn, TikTok, Instagram and YouTube. Every feature is on every plan, you see the cost before anything renders, and a failed render is never charged.

Rendering runs in the background in the cloud; you can upload another video while one renders, and if you disconnect, the job resumes where it stopped.

Upload one raw video and see the finished edit.
Free plan: 5 watermarked videos a month. No credit card.

See your first video — free

Claude Code + Remotion vs RealTake

RealTakeClaude Code + Remotion
Where it rendersIn the cloud, in the backgroundOn your machine, 2 to 10 minutes a pass
Two at onceYes; upload the next while one rendersReported glitching and breaking midway
On-screen text and captionsWord-level captions, titles and 200+ motion graphics placed clear of your facedrawtext "unreliable"; captions depend on libass being installed
If it failsNever charged; the job resumes if you disconnectYou debug the stack
InstallNothingFFmpeg, Remotion, a caption model, their dependencies

Do-it-yourself figures from our own tests and the published workflows linked on the page, read October 28, 2026.

FAQs

Why do my Remotion renders glitch when I run several?
Published write-ups report the same thing: several renders competing on one machine tear the output and break midway. RealTake renders in the cloud and lets you queue the next video while one renders.
Why does FFmpeg say "No such filter: 'subtitles'"?
One write-up traces it to a Homebrew FFmpeg built without libass. RealTake needs no FFmpeg; captions are generated word-level and placed on the frame for you.
Does RealTake need anything installed?
No. It runs in the browser on a phone or laptop; rendering happens in the cloud.

Your next video is already on your phone.

Upload it and see it fully edited in minutes.

See your first video — freeFree plan: 5 watermarked videos a month · no credit card