Cost to serve 1 credit = 1 minute of source video: Tornado / Vitruvian download + spot CPU workers (c7i, eu-west-1) + LLM + R2 + Paystack. Defaults are the live pack prices. See sources & verification at the bottom.
30m = 0.90 GB
Pricing page: "as low as $0.06/GB" on pay as you go and custom plans. The actual rate shows under Billing in your dashboard. A billing floor can charge short or audio-only jobs at the source size.
Unverified: no Vitruvian pricing page could be found. $2.00/GB is a placeholder. Paste their per-GB rate here.
No YouTube captions: ASR runs on the worker
CPU s per source min (queue.py default)
CPU s per source min (queue.py default)
Fixed work per job (queue.py default)
ASG scales from 0: boot takes 2-3 min
Spreads overhead + cold start
Spot reclaims + failed jobs (no revenue)
-
-
-
-
Where every Naira goes per 1 minute of generated video before profit margin.
| 1. Tier Pricing & Allocation (Editable) | 2. Shared Infra Stack | 3. All-In Tornado | 4. All-In Vitruvian | 5. Tornado Advantage | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Tier Name | Price (₦) | Credits (Mins) | Rate/Min & Paystack | Spot + LLM + Stack | All-In Cost ($ / ₦) | Net Profit & Margin | All-In Cost ($ / ₦) | Net Profit & Margin | Saved w/ Tornado | |
Margin badges highlight profitability after deducting all costs:
Revenue -
Tornado cost -
Vitruvian cost -
Revenue -
Tornado cost -
Vitruvian cost -
Revenue -
Tornado cost -
Vitruvian cost -
Margin in brackets. Revenue uses the average ₦/credit across the tiers above. Pricing modeled as 2K = 3x and 4K = 4x credits. Note: the backend currently charges 4x for both (CLIPPING_HIGH_RES_MULTIPLIER) and sabiclip.com says "2K / 4K = 4x", so both need changing to match. Download, CPU and storage scale with the file-size multiplier (estimate). LLM cost does not change with resolution.
Checked November 2026 against the SabiClip code and public price pages. "Verified" means a public source or the code confirms it. "Estimate" means it depends on our own usage and should be measured from real jobs.
| Item | Used here | Status | What we found |
|---|---|---|---|
| Tornado API | $0.06/GB (≈ $0.0018/min) | Verified (pricing page) | Pricing page, checked today: "as low as $0.06 per GB" on pay as you go, with the lowest rate tied to a volume commitment. Custom plans also start at $0.06/GB for committed volume above 30 TB. Billable volume is the delivered file size unless a billing floor applies. Failed jobs are never billed. 100 GB free for 7 days on new accounts. Used as the first-line YouTube downloader in tornado_ingestion.py. Confirm your account rate under Billing. tornadoapi.io/pricing |
| Vitruvian API | $2.00/GB placeholder | Not found | Searched for a Vitruvian pricing page and tried the likely domains (vitruvianapi.com/.io/.dev, vitruvian-api.com). None returned a page, and the name is not in the codebase or env. The input is per GB so their rate can go straight in. $2.00/GB equals the old $0.06/min placeholder at 30 MB/min. |
| Spot workers | $0.11/hr | Verified | The render ASG (infra/terraform/render-workers) is CPU only: c7i.xlarge/2xlarge, c6i.xlarge/2xlarge, c5.xlarge, all spot, in eu-west-1, scaling 0 to 20. c7i.xlarge spot is about $0.10 to $0.11/hr and on-demand about $0.19/hr, with 15 to 20% interruption risk. DoiT price tracker |
| CPU seconds per source minute | 12 s captions / 55 s Whisper, +45 s per job | Code defaults | Taken from apps/clipping/queue.py (_DEFAULT_RATE, _DEFAULT_OVERHEAD), which the app uses for ETAs until it has history. The app replaces them with the median of the last 60 finished jobs, so use those once real jobs exist. Whisper share (30%) and cold start billed per job (1 min) are guesses. |
| Whisper (ASR) | No separate cost (hosted API off) | Verified in code | Default engine is local faster-whisper (small, int8, CPU) on the same worker, so its cost is the extra CPU seconds above. It runs only when the video has no captions. The OpenAI Whisper API is used only if a site sets asr=openai, so that line is off. If enabled, whisper-1 list price is $0.006/min; check OpenAI pricing. |
| LLM | $0.0003/min | Estimate | Two OpenAI chat calls per job (find moments, fix transcript), default model gpt-4o-mini. About 500 input and 250 output tokens per source minute at roughly $0.15 / $0.60 per million tokens gives about $0.0002 to $0.0003/min. A site set to gpt-4o would be about $0.003/min. Check the model set in your AI provider config. |
| Storage (R2) | $0.0005/min | Estimate | R2 stores at about $0.015 per GB-month with no egress fee. A 30 MB source minute held a month is about $0.0005. Lower if sources are deleted sooner. |
| TTS | Not included | Removed | The clipping pipeline has no TTS. It belongs to Video Studio, which has different costs. |
| Paystack | 1.5% + ₦100, cap ₦2,000 | Verified | The live provider (PAYMENT_PROVIDER defaults to paystack; DEPLOYMENT.md still mentions bachs and is out of date). ₦100 waived under ₦2,500. Applied once per top-up, not per job. paystack.com/pricing |
| Credits and prices | Live pack prices | From site and code | 1 credit = 1 source minute, rounded up per job, held at start, charged on success, released on failure. Page models 2K = 3x and 4K = 4x credits. The code currently charges 4x for both (CLIPPING_HIGH_RES_MULTIPLIER). Pack prices are from sabiclip.com. The "Draft tiers" button loads the earlier proposal. Unused credits never expire, so breakage is not modeled. |
| Not included | - | Gap | Fixed costs: app box (t3.small about $15/mo plus about $2.40 disk, per DEPLOYMENT.md), Postgres/Redis, and Cloudflare. Also support, refunds, and failed jobs beyond the waste %. |
| FX rate | ₦1,600/$ | Assumption | Matches USD_TO_NGN in the backend settings. Every Naira figure scales with it, so test 1,400 to 1,700. |