What $52 in API costs actually produces

We track every API call our pipeline makes. Every token in, every token out, every cent spent. Here are the real numbers.

Two and a half cents per finished video. That includes arc extraction, title generation, SEO metadata, and a full quality review by seven AI agents. Let me break down where the money actually goes.

Where the money goes

Arc extraction is the most expensive step because it's the most demanding. The model reads a full stream transcript — sometimes 50,000+ words — and has to understand narrative structure, identify where stories begin and end, and decide what's worth extracting. That requires a large context window and careful reasoning. Worth every fraction of a cent.

What's free (no API cost)

Most of the pipeline doesn't touch an API at all.

• Whisper transcription — runs locally on an RTX 6000 GPU. Full speech-to-text for an 8-hour stream in about 20 minutes. Zero API cost.
• FFmpeg assembly — all video cutting, encoding, thumbnail generation, and audio normalization runs on CPU. Zero API cost.
• YouTube upload — the YouTube Data API has a daily quota, not a per-call charge. Uploading 20 videos costs the same as uploading 1. Zero API cost.

The expensive parts of video production — transcription and rendering — are all local compute. The API costs are only for the intelligence layer: understanding content, generating metadata, and quality control.

The comparison that matters

A freelance video editor charges $50-200 per finished video. That's the market rate for someone who watches your VOD, finds the good parts, cuts a video, writes a title, and delivers a final product.

At $0.025 per video, we produce 2,000 videos for what a freelancer charges for one.

That's not a typo. Two thousand videos for the cost of one freelance edit.

Now — is every AI-produced video as good as what a skilled human editor would make? No. A talented editor with deep knowledge of a creator's brand will produce better individual videos. That's not the comparison that matters.

The comparison that matters is: what does a streamer who can't afford $50-200 per video actually do? The answer, for most streamers, is nothing. They don't repurpose their content at all. Their streams expire after 14 days on Twitch and disappear forever.

$0.025 per video means the cost barrier is gone. The question shifts from "can I afford to repurpose?" to "how do I want to repurpose?"

The catch

Not all videos are equal. The cost-per-video is an average, and the quality varies by format.

Full story arcs cost more per video because the extraction step requires more reasoning over longer content windows. But they pass review at 89% — nearly 9 out of 10 make the cut. The content is richer, the narratives are more complete, and the quality review has more to work with.

Shorts are cheaper to produce but harder to get right. A 30-second clip needs to hook instantly and deliver a complete moment. At a 54% pass rate, almost half get killed by the review panel. The ones that survive are genuinely good. The ones that don't would have been scroll-past content that wastes a viewer's time and an upload slot.

We'd rather produce 100 Shorts, kill 46 of them, and upload 54 bangers than upload 100 mediocre clips and let the algorithm punish the channel.

What this means for creators

The economics of content repurposing have fundamentally changed. When the marginal cost of producing a video is two and a half cents, the bottleneck isn't money — it's quality control. That's why we spend more of our API budget on review (the 7-agent panel) than on any other single stage.

Cheap production without quality gates produces spam. Cheap production with aggressive quality gates produces a content library. We're building the latter.

Plans start at $149/mo. That's about 4,000 videos worth of API costs.