Every AI video tool shows you a demo. Almost none of them show you the invoice. We produce full vertical-drama episodes (17+ cuts each) and webtoon seasons through our own pipeline, so we track generation cost per cut the way a studio tracks film stock. This post shares the numbers we actually verified in August 2026 — the same shot, priced across every route we could call or click.
The benchmark shot
One 5-second, 720p, image-to-video generation on Seedance 2.0 — the workhorse setup for short-drama cuts: you generate a keyframe first, then animate it. Prices below are from published rate cards and live pricing pages, checked on the same day.
What the same cut costs, by route
| Route | Price per 5s cut | Notes | | --- | --- | --- | | Dreamina subscription (annual) | ~$0.51 | Consumer app, ~101 generations/month cap, manual only — no API | | Poyo API (fast tier) | ~$0.70 | Cheapest documented API rate we found | | Poyo API (standard) | ~$1.00 | Per-second billing, no subscription | | EvoLink API (standard) | ~$1.00 | Comparable to Poyo standard | | fal API (standard) | ~$1.51 | Official partner pricing, ~50% above the cheapest APIs |
Three things jump out:
- The consumer subscription is half the price of any API. That is not the API being greedy — it is breakage economics. Most subscribers never use their full monthly quota, and the unused credits subsidize the ones who do. If you only use half your quota, your effective price doubles and the advantage disappears.
- APIs vary by ~2x for the identical model. Same weights, same output — the difference is pure distribution margin. If your tool locks you to one provider, you are paying that spread without knowing it.
- "Unlimited" plans are speed-capped, not volume-capped. Time-boxed unlimited offers are real, but queue priority is the actual limit — you cannot batch an episode through a web UI at unlimited speed, and automation surfaces are excluded from those plans.
What an episode costs, not a cut
A cut is not the unit of production — an episode is. A one-minute vertical-drama episode at proper cut density (25–30 cuts per minute is what hit shows actually run) needs 25+ video generations, plus keyframes, plus retakes. Our real-world retake rate on acting-heavy cuts runs 2–4 takes per keeper. So a one-minute episode lands roughly here:
- Keyframes (30 images): $1–3
- Video cuts (25–30 keepers at 1–3 takes each): $25–90 via API
- Voice, music, subtitles: $1–5
Call it $30–100 per finished minute on paid APIs — and near the bottom of that range only if your pipeline avoids re-billing for identical inputs and routes each cut to the cheapest model that can handle it.
How we keep it at the bottom of the range
This is exactly why we built owntent around cost routing instead of a single model:
- BYOK by default. Your API keys, provider cost passed straight through, no markup on generation.
- Per-cut model routing. Acting-heavy cuts go to the strongest model; establishing shots and drafts go to cheaper tiers or a local GPU at marginal cost zero.
- Cost shown before you commit. Every generation displays its credit cost up front, from the same rate table our billing actually uses.
- No re-billing on identical inputs. Hash-cached generations are free to reuse; a script tweak only regenerates the cuts it touches.
The cheapest cut is the one you do not regenerate blindly. Tools should make cost a visible, routable variable — not a surprise at the end of the month.
Rates checked 2026-08-05 from public rate cards (fal, Poyo, EvoLink) and in-app pricing (Dreamina, Higgsfield). Prices change fast in this market — treat this as a method, not gospel.