Contains affiliate links. We may earn a commission — it never changes the verdict.Details
Guides / updated 2026-07-03
The AI YouTube Shorts Pipeline: Script to Published in Under an Hour
A production-grade AI shorts pipeline is five seats: Claude writes the script, HeyGen turns it into a talking-head video, CapCut adds captions and polish, and a scheduler ships it — with one compliance step wired in that most creators skip and regret. This is the pipeline we actually run for the Stackaible channel. Total tool cost: roughly $75/month.
The pipeline
| Stage | Tool | Job | Cost |
|---|---|---|---|
| Script | Claude | 135-word script that answers the title in sentence one | $20/mo |
| Presenter | HeyGen | Custom avatar + cloned voice reads the script | ~$29–59/mo |
| Edit | CapCut | Captions, pacing, brand overlays | Free |
| Distribute | Buffer | Queue across platforms | Free–$6/mo |
| (Optional) Repurpose | OpusClip | Long-form → shorts candidates | $15/mo |
Stage 1: The script is the product
Everything downstream executes what the script decides. Two rules that matter more than any tool:
- Sentence one answers the title. The thumbnail made a promise; pay it off in the first breath. No throat-clearing.
- ~2.3 words per second. A 60-second short is 135–140 words. Overwriting is the most common pipeline failure and it happens before any AI touches video.
Prompt Claude with your topic, your take, and those two constraints. Reject drafts that warm up before they answer.
Stage 2: The presenter
HeyGen with a custom avatar and your cloned voice turns the approved script into a video in minutes. Invest the setup time in your own avatar — stock avatars read as stock, and the whole game is looking current. Budget honestly: ~$60/month once priority processing is real usage, not the $29 sticker.
If your format is narration-over-visuals instead of talking-head, swap HeyGen for ElevenLabs voice (from $6/month — the tier that includes the commercial license a monetized channel needs).
Stage 3: The finish
CapCut free tier: auto-captions (always — most shorts are watched muted), a pacing pass to kill dead air, and your brand overlays. This stage is 15 minutes, not an afternoon; polish has diminishing returns at 9:16.
The compliance step — do not skip
If the video promotes a product you have an affiliate relationship with, the FTC requires disclosure verbally and on-screen, in the first moments of the video — a description-only disclosure is a violation, and fines run to five figures per post. Also flip YouTube’s “Includes paid promotion” toggle; it’s a platform requirement, not a substitute for in-video disclosure. Wire both into your template so compliance is structural, not remembered.
The loop
Script Monday, render Tuesday, publish on queue all week, review retention data Friday, feed what worked back into next Monday’s scripts. The pipeline is the machine; the loop is what makes it compound.
Tools in this guide
Claude
Chat AssistantsThe thinking seat: long documents, hard reasoning, code, and anything where the words have to be right the first time.
HeyGen
VideoThe talking-head seat: avatar videos from script, lip-synced translation, and presence at scale without a camera setup.
ElevenLabs
Voice & AudioThe voice seat: narration, voice cloning for scaled content, and audio versions of written work.
CapCut
VideoThe vertical-edit seat: cutting, captioning, and shipping 9:16 video at the speed shorts demand.
OpusClip
VideoThe repurposing seat: turns long-form video into scored, captioned vertical clips ready for Shorts, Reels, and TikTok.