Content Studio
Ten niche story channels run by one agent pipeline. The second attempt, built after the first one taught me what was wrong.
- 10
- channels, each a config entry and not a fork
- 8
- milestones planned and built
- 0
- lines of code carried over from the first version
- 0
- paid services until the channels earn
01Summary
Content Studio turns Reddit stories into narrated short videos and manages ten separate channels, one niche each, from a single dashboard. Each channel is the same pipeline with a different configuration: sources, voice, theme and a daily upload cap.
It is a ground-up rebuild of my earlier project, StoriqueOS. Nothing was ported.
02Why I built it
The first version worked and I did not like what it made. The voices sounded robotic. The scripts read strangely. It pushed seven to ten videos a day through one channel, which invites a shadow ban. And sourcing was a flat scrape with no judgement in it. Patching it would have kept the wrong shape.
03How it works
- Step 1
Source
Read stories through public feeds, with no credentials.
- Step 2
Scan
Look at what other creators in the niche are doing before writing any metadata.
- Step 3
Script
Rewrite the story for narration.
- Step 4
Voice
Synthesise speech with a local voice model.
- Step 5
Assemble
Background footage, voice, an opening story card, then word-level captions.
- Step 6
Learn
Feed performance back into what gets picked next.
04Decisions
| Decision | Reason |
|---|---|
| Fewer uploads per channel | The shadow ban risk came from volume, so volume went down and channels went up. |
| A real agent framework | Ten channels running at once exposed concurrency problems a hand-rolled pipeline would have hidden. |
| Licensed footage only | Creative Commons and free stock sources, not arbitrary scraping. |
| Verified is a separate word from wired | The documentation lists what was proven with real data and what only degrades gracefully until credentials are added. |
05What broke
- The first version, as a whole. That is the lesson: I rebuilt it because the output was not good enough, not because the code failed.
- Running all ten channels for real exposed concurrency bugs that a single channel never showed.
- Without credentials every agent returns empty. I tested that it does so without crashing.
06What is next
Publishing credentials for each channel, then the two other content systems on the same framework.