Yes, this automates 25-minute videos too: my long-form space channel pipeline
Every post so far covered 15-20 second shorts. Fair question: does the draft-file approach survive contact with real long-form? My English space channel says yes: six 25-minute documentaries have shipped through a long-form variant of the same pipeline.
What changes at 25 minutes
The core trick is identical (write CapCut draft JSON directly), but long-form breaks three assumptions the shorts pipeline makes:
- Scene count explodes. A short has 3-5 scenes; a documentary has dozens of segments, each with its own visuals, narration block, and subtitle run. The build script assembles them from a structured script file instead of a flat text blob.
- Subtitles become a subsystem. At 25 minutes you’re placing hundreds of caption lines. Hand-timing them would take longer than the entire rest of production. Measured-audio timing (same as shorts) is what makes long-form viable at all.
- Visuals are the real cost. The narration and assembly are near free; the AI-generated space visuals are where actual money goes. That subscription is the single paid item in my whole stack, and it belongs to this channel, not to the editing pipeline.
The receipts
| Item | Number | Source |
|---|---|---|
| Long-form docs shipped | 6 | YouTube Studio |
| Runtime each | ~25 min | channel |
| Shorts recut from them | 18 | next post in this series |
| Paid tools in the loop | 1 (AI visuals) | my billing |
The experiment that failed
Honesty section. I ran a thumbnail-swap experiment on the first documentary to try to lift click-through:
| Metric | Before | After | Verdict |
|---|---|---|---|
| CTR | 2.5% (321 impressions) | 2.4% (294 impressions) | no effect |
Target was 4%. The swap did nothing measurable, and the impression counts are small enough that even “nothing” is uncertain. I’m publishing it anyway because a receipts site that hides flat results is lying by omission. Long-form growth on this channel is unsolved. The pipeline makes episodes cheap to produce; it does not make them get clicked.
What you can use today
- If you script long-form, structure the script by segment from day one (even just headers in a text file). Every automation and every edit gets easier downstream.
- Generate narration before editing and cut to audio, exactly like shorts. At 25 minutes this saves the most time of anything here.
- Before spending on thumbnails or tools, check your impression count. Under a few thousand impressions, you can’t measure whether anything worked. I learned that from the table above.
Here are the receipts.