# turns out the AI wasn't my expensive part, rendering was — retained record

> Retained reddit record from reddit:SaaS.

Canonical: https://highsignal.app/data/records/61600d40dd8a2896

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:SaaS)  published 2026-10-03 11:06 ingested 2026-10-04 02:37

[original source ↗](https://www.reddit.com/r/SaaS/comments/1wwkkge/turns_out_the_ai_wasnt_my_expensive_part/)

## retained text



```
solo dev here, building an app that generates short videos. when I started I was sure the LLM calls would be my biggest cost and headache. wrong.

the model writes each video as an html page, then I render it frame by frame with headless chrome and ffmpeg. on my laptop one video takes 1 to 5 min. on a small cloud container, 3 renders at the same time just hang forever.

 so now I'm looking at a proper render queue with a concurrency limit instead of firing everything at once.

anyone here running heavy jobs like this (video, pdf, image processing)? do you queue on your own servers or push it to something like a job service? and how did you price it so the compute doesn't eat the margin?

WDYT 
```



document fetched 2026-10-04 00:27 · canonical https://reddit.com/r/SaaS/comments/1wwkkge/turns_out_the_ai_wasnt_my_expensive_part
