tony677
Newbie
- Feb 10, 2026
- 23
- 15
Been reading journey threads here for a long time, figured it's finally my turn to start one.
The short version: I'm a dev. I got tired of manually clipping streams, so over the past few
months I built my own pipeline that does the whole thing end to end. It watches for new stream
VODs, scans the full video with AI to find the moments actually worth clipping, cuts them,
crops to vertical, burns in subtitles, and posts them on a schedule. no editors, no VAs,
no me sitting in CapCut at 2am.
Now I'm pointing it at two channels and documenting how far it can go.
THE CHANNELS
- YouTube Shorts: chess niche (clips from chess streamers)
- TikTok: CS2 (clips from Kick streams)
Why chess? Everyone runs gaming clip channels, almost nobody automates chess. The audience
is older, watch-through is high, and the supply of streamer content is endless. CS2 is the
control group - the obvious niche - so I can compare a saturated niche vs an underserved one
with the exact same pipeline.
THE SETUP
1. Watcher polls a list of streamers for new VODs
2. AI pass scans the whole VOD and scores moments on hook strength - whether the first 1-2
seconds would stop a scroll. This took the longest to get right and it's where 90% of
clip channels fail imo
3. Best moments get cut, auto-cropped to vertical, subtitled
4. Scheduler posts with 12h (+-30min) gaps between uploads so the channels don't look botted
Everything runs on a cloud server. I check the dashboard maybe 10 minutes a day. Some days
I don't look at it at all and clips still go out.
WHERE THINGS STAND TODAY
Small numbers right now as u can see in screenshots. That's the point of the thread anyways. if I waited until it was impressive
there'd be nothing to follow.
THE GOAL
10M combined views. Checkpoints I'm holding myself to:
- 100K combined - prove the pipeline works at all
- 1M combined - prove a niche, not luck
- 10M - prove it scales by adding channels, not hours
I'll post updates roughly weekly with real screenshots, including the flops. ask me anything
)
about the setup - worst case someone here finds a hole in my method, which is also useful.
The short version: I'm a dev. I got tired of manually clipping streams, so over the past few
months I built my own pipeline that does the whole thing end to end. It watches for new stream
VODs, scans the full video with AI to find the moments actually worth clipping, cuts them,
crops to vertical, burns in subtitles, and posts them on a schedule. no editors, no VAs,
no me sitting in CapCut at 2am.
Now I'm pointing it at two channels and documenting how far it can go.
THE CHANNELS
- YouTube Shorts: chess niche (clips from chess streamers)
- TikTok: CS2 (clips from Kick streams)
Why chess? Everyone runs gaming clip channels, almost nobody automates chess. The audience
is older, watch-through is high, and the supply of streamer content is endless. CS2 is the
control group - the obvious niche - so I can compare a saturated niche vs an underserved one
with the exact same pipeline.
THE SETUP
1. Watcher polls a list of streamers for new VODs
2. AI pass scans the whole VOD and scores moments on hook strength - whether the first 1-2
seconds would stop a scroll. This took the longest to get right and it's where 90% of
clip channels fail imo
3. Best moments get cut, auto-cropped to vertical, subtitled
4. Scheduler posts with 12h (+-30min) gaps between uploads so the channels don't look botted
Everything runs on a cloud server. I check the dashboard maybe 10 minutes a day. Some days
I don't look at it at all and clips still go out.
WHERE THINGS STAND TODAY
Small numbers right now as u can see in screenshots. That's the point of the thread anyways. if I waited until it was impressive
there'd be nothing to follow.
THE GOAL
10M combined views. Checkpoints I'm holding myself to:
- 100K combined - prove the pipeline works at all
- 1M combined - prove a niche, not luck
- 10M - prove it scales by adding channels, not hours
I'll post updates roughly weekly with real screenshots, including the flops. ask me anything
)about the setup - worst case someone here finds a hole in my method, which is also useful.
