Uzair SaleemDesign&Desktop
AI reel generator
An AI reel generator for Design&Desktop, a team in Germany. I built the tool and shipped the n8n workflow that posts the finished reels.
- Specscenes, text, media
- RenderRemotion
- EncodeMediaBunny, 10 sec
- Publishn8n to Instagram
The problem
Rendering ran in the user's browser. The same clip took 2 minutes on one device and 10 on another, so there was no honest render time to promise anyone.
What I built
Reels from a spec
Remotion compositions render each reel from a spec of scenes, text and media, so a new reel is data, not a new edit.
Posting that fails loudly
n8n handles credentials, retries and the posting schedule, so one broken API call doesn't quietly skip a day.
The hard part
Making renders fast on every device
The first version used FFmpeg compiled to WebAssembly. Multithreading it cut the time about in half, which was not enough. I looked at the alternatives and moved the pipeline to MediaBunny, which uses the browser's own video encoders through WebCodecs. A clip now takes about 10 seconds.
Results
- Render time went from 2 to 10 minutes down to about 10 seconds per clip.
- Finished reels post to Instagram on a schedule with no one opening a video editor.
Also at Design&Desktop
- Restaurant data dashboard: parallel headless Chromium workers made scraping 80% faster.
- German B2B marketplace: delivery pricing and address checks across 12,000 postal codes.
Stack
Contact
Want me on your team?
uzair@uzairsaleem.dev- Role
- Full-time remote. Full-stack, backend or product engineer.
- Hours
- US hours from Islamabad (UTC+5).
- Paperwork
- Employee or contractor. Deel, Payoneer or a wire all work.
- Start
- Available now.