Direct answer: AI video pricing in 2026 can be reduced to a per-second number, but the invoice rarely grows the way you expect. Using Runway's public prices (1 credit = USD 0.01), an 8-second clip ranges from USD 0.16 (Grok Imagine Lite 480p) to USD 3.20 (Veo 3.1 with audio). A 30-second spot built from 4 shots with 3 attempts each costs USD 4.80–38.40 in generation alone, before editing, voice-over and licensing. What actually drives the budget is retry rate, billing block size, whether input video is charged, and 4K/HDR multipliers — not the headline per-second price.
This guide is for marketing, e-commerce, content and procurement teams producing product films, ads and social clips. Data cut-off: 31 October 2026; prices come from official Runway, Google Cloud and Luma pages, in USD, excluding tax, with no FX conversion and no ranking. The "cost per usable clip" and "30-second spot" figures are editorial models, not quotes.
Convert everything to a per-second price before comparing; cost per usable clip = clip price × attempts; 4K, HDR, input video and minimum charges are the four main reasons invoices exceed the unit-price estimate.
Why can't you compare "price per video"?
Because vendors bill in different units: Runway per second of credit, Luma per 5-second block, Google per count, and some models add a minimum charge. Only after converting to one unit can an 8-second clip and a 30-second spot be compared.
| Vendor / page | Official billing unit | What it includes | Common trap |
|---|---|---|---|
| Runway Dev | Credits per second; 1 credit = USD 0.01 | Billed on output seconds; audio toggle, resolution and model tier | Input video or reference images may be billed separately (Kling 4, Seedance 2.5, Hailuo 3) |
| Luma Ray3.2 | Per 5-second block; SDR price, HDR 2×, HDR+EXR 3× | Generation, edit and Reframe priced separately | 5-second and 10-second list prices are not a simple multiple; confirm in console |
| Google Veo 3.1 | Per 1 count; tiered by resolution and audio | 4/6/8-second clips, up to 4 outputs per prompt | The count-to-second conversion is not on the pricing page |
| Seedance 2.0 | Per second, with a 56-credit minimum per generation | Tiered by output resolution | Short clips still pay the minimum |
| Kling 3.0 | Per second; audio and no-audio are different prices | 4K tier priced separately | Kling O3 bills the same whether audio is on or off |
Sources: Runway "API Pricing & Costs", Luma "API Pricing", Google Cloud Vertex AI pricing and Veo 3.1 documentation. Only units verified in official pages are listed; missing models do not mean cheaper options do not exist.
What do per-second prices mean for 8, 30 and 60 seconds?
At Runway list prices: Veo 3.1 with audio is USD 0.40/s — USD 3.20 for 8 seconds, 12 for 30, 24 for 60. Gen-4 Turbo at USD 0.05/s costs 0.40/1.50/3.00 for the same durations.
| Model (Runway price) | USD/second | 8 seconds | 30 seconds | 60 seconds |
|---|---|---|---|---|
| Veo 3.1 (with audio) | 0.40 | 3.20 | 12.00 | 24.00 |
| Veo 3.1 Fast (with audio) | 0.15 | 1.20 | 4.50 | 9.00 |
| Kling 3.0 Standard (with audio) | 0.13 | 1.04 | 3.90 | 7.80 |
| Gen-4.5 | 0.12 | 0.96 | 3.60 | 7.20 |
| Wan 3.0 (720p) | 0.10 | 0.80 | 3.00 | 6.00 |
| Gen-4 Turbo | 0.05 | 0.40 | 1.50 | 3.00 |
| Grok Imagine 1.5 Lite (480p) | 0.02 | 0.16 | 0.60 | 1.20 |
| Seedance 2.0 (720p) | 0.30 | 2.40 | 9.00 | 18.00 |
Editorial conversion: credits/second × USD 0.01 = USD/second; × 8/30/60 gives the examples. Excludes tax, retries, reference images/input video, 4K/HDR surcharges and labour. Runway's Kling 4, Seedance and other models may also carry promotional prices or minimum charges; the invoice governs.
Is Google's "count" really one second?
Google's pricing page lists "per 1 count" without defining the conversion, but Runway's resale price for Veo 3.1 matches Google's count price exactly in the flagship tiers (USD 0.40 with audio, USD 0.20 without), which supports "1 count ≈ 1 second" — still to be confirmed in the console.
| Veo tier | Google official price | Runway resale (USD/second) | Aligned? |
|---|---|---|---|
| Veo 3.1 with audio, 720p/1080p | USD 0.40 / 1 count | 40 credits/s = 0.40 | Exact match |
| Veo 3.1 without audio, 720p/1080p | USD 0.20 / 1 count | 20 credits/s = 0.20 | Exact match |
| Veo 3.1 with audio, 4K | USD 0.60 / 1 count | No separate 4K tier | Cannot cross-check |
| Veo 3.1 Fast with audio, 720p | USD 0.10 / 1 count | 15 credits/s = 0.15 | Mismatch, possibly resale margin |
| Veo 3.1 Fast with audio, 1080p | USD 0.12 / 1 count | 15 credits/s = 0.15 | Close, still above official |
| Veo 3.1 Fast without audio | USD 0.08–0.10 / 1 count | 10 credits/s = 0.10 | Matches the 1080p tier |
Google prices from the Vertex AI pricing Veo table; Runway prices from its per-second credit table. The flagship match is circumstantial evidence for "1 count = 1 second", but this article still labels it an editorial inference; the Fast-tier gap shows resale price is not always the official price, so direct and resale purchases must be checked separately. If count = 1 second, Veo 3.1 with audio is USD 3.20 for 8 seconds and 12 for 30; if count means "per clip", the estimate is off by several times. Run one 4-second and one 8-second job and reconcile the invoice before signing.
Luma uses another unit: Ray3.2 bills per 5-second block, with 720p generation at USD 0.30 for 5 seconds and 1080p at USD 1.20; HDR output is 2× and HDR+EXR 3×. Its 10-second list price (720p, USD 0.90) is not simply twice the 5-second price, while its FAQ says billing is per 5-second block — rather than argue, run a 10-second job and read the invoice.
Retries are the real cost multiplier
One 8-second Veo 3.1 clip with audio is USD 3.20; if it takes 3 attempts to get one usable clip, the usable-clip cost is USD 9.60. Cheap generation is not cheap usable footage.
| Attempts | Veo 3.1 8s (0.40/s) | Veo 3.1 4K 8s (0.60/s) | Gen-4 Turbo 8s (0.05/s) |
|---|---|---|---|
| 1 (usable first time) | USD 3.20 | USD 4.80 | USD 0.40 |
| 2 | USD 6.40 | USD 9.60 | USD 0.80 |
| 3 | USD 9.60 | USD 14.40 | USD 1.20 |
| 5 | USD 16.00 | USD 24.00 | USD 2.00 |
Formula: cost per usable clip = clip price × attempts. Replace attempts with your own historical rate; the table excludes prompt iteration, human selection and editing.
What does a 30-second spot actually cost?
Assuming 4 shots × 8 seconds with 3 attempts each — 96 seconds generated — Veo 3.1 with audio costs USD 38.40, Gen-4 Turbo 4.80, Wan 3.0 720p 9.60 and Seedance 2.0 720p 28.80.
| Option | USD/second | 96 seconds generated | Typical use |
|---|---|---|---|
| Veo 3.1 (with audio) | 0.40 | USD 38.40 | Brand films needing synced dialogue/SFX |
| Seedance 2.0 720p | 0.30 | USD 28.80 | High-resolution social clips; watch the minimum charge |
| Veo 3.1 Fast (with audio) | 0.15 | USD 14.40 | Fast iteration and social variants |
| Gen-4.5 | 0.12 | USD 11.52 | General shots, audio added later |
| Wan 3.0 720p | 0.10 | USD 9.60 | Budget-sensitive bulk footage |
| Gen-4 Turbo | 0.05 | USD 4.80 | Drafts and storyboard previews |
Editorial model: 4 shots × 8 seconds = 32 seconds of finished footage, 3 attempts = 96 seconds generated; excludes editing, voice-over, music licensing, subtitles, storage and human review. If the deliverable needs 4K or HDR, apply the official multipliers (e.g. Luma HDR 2×, HDR+EXR 3×; Veo 4K USD 0.60/s).
Why does the invoice exceed the unit-price estimate?
Four main reasons: retries, separately billed input video/reference images, minimum charges, and 4K/HDR/professional-format surcharges. Any one can turn "a few cents per second" into two or three times that.
| Surcharge | Official example | Budget impact |
|---|---|---|
| Input video | Seedance 2.5: 20/30/68 credits per output second, plus 10/15/34 per input/reference second | Video-to-video can approach double |
| Reference images/keyframes | Hailuo 3: 2 credits per reference image or keyframe; Grok 1.5: 1 credit each | Multiple references add up per image |
| Minimum charge | Seedance 2.0: 56 credits minimum per generation; Seedance 2.5: 80 credits | 1–2-second clips get more expensive per second |
| Professional formats | Runway ProRes/PNG sequence +5 credits/s; HDR +20 credits/s (40 above ~4MP) | Master delivery adds a lot |
| HDR/EXR | Luma: HDR 2×, HDR+EXR 3× the SDR price | Cinema delivery is priced as a multiple |
| Audio toggle | Kling 3.0: 9 credits/s without audio, 13 with | Budget dialogue films at the audio tier |
Official examples from Runway, Luma and Google documentation; actual invoices may also reflect promotions (e.g. Kling 4 at 20% off through 30 November 2026), region and tax.
What should procurement and acceptance check?
Compare on one unit, confirm block size and minimum charges, budget for retries, fix resolution and audio deliverables, and check commercial and likeness rights — not just the per-second number.
| Acceptance item | Suggested definition | Evidence to keep |
|---|---|---|
| Billing unit | Convert to USD/second, state block size and minimum charge | Official price screenshots, sample invoice |
| Retry budget | Based on historical rate, or 3 attempts per shot | Generation log, usable-rate stats |
| Input assets | Confirm whether image-to-video and video-to-video are charged separately | Request parameters, invoice line items |
| Delivery spec | Resolution, frame rate, audio, HDR/EXR surcharges | Sample clips, master spec |
| Rights | Commercial licence, likeness, brand elements | Terms, licence documents |
| Labelling/watermark | AI-generated labels, removable platform watermark | Final files, platform rules |
| Concurrency/queue | Peak turnaround, whether failed retries are billed | Load test, SLA |
| Exit/migration | Model shutdown, asset export, alternatives | Contract clause, export drill |
Editorial guidance, not legal advice; commercial licensing, likeness and labelling rules must be checked per publishing region.
Who should read this — and who should not?
For teams producing bulk short video, product films or ad creative. Not for one-off experiments, or projects treating AI video as a zero-cost asset library.
Read if: you are in marketing, e-commerce, content or procurement; you are comparing Runway, Veo, Kling or Seedance APIs; you need a budget or acceptance basis for a 30-second spot. Skip if: you only run a one-off proof of concept; you need live-action talent; you want this article to make creative decisions for you. Supports: like-for-like comparison, retry budgeting, finished-spot costing, delivery specs and contract acceptance.
Lingyi does not rank video models or promise output quality. To centralise multi-model calls, quotas, invoice aggregation and fallback switching, see LingHarness; to fold video cost into project budgets and acceptance, talk to us.
Related reading: AI image cost and usable rate, vendor shutdown and migration cost, and AI voice-over and licensing cost; this article only covers per-second/per-block video pricing and finished-spot modelling.
FAQ
Is AI video billed per second or per clip?
Both. Runway bills credits per second (1 credit = USD 0.01), Luma bills per 5-second block, Google Veo bills per 1 count, and some models add a minimum charge per generation. Convert to USD/second and state the block size and minimum before comparing.
What is the cheapest 8-second video?
Among the public prices verified here, Grok Imagine 1.5 Lite 480p on Runway is USD 0.02/s, or 0.16 for 8 seconds; Gen-4 Turbo is 0.40; Wan 3.0 720p 0.80; Veo 3.1 with audio 3.20. Cheaper tiers usually trade resolution, duration or consistency, so match the tier to the use case.
Is Veo's count one second?
Google's pricing page only says "per 1 count" without defining the conversion. However, Runway's resale price for Veo 3.1 matches Google's count price exactly in the flagship tiers (USD 0.40 with audio, 0.20 without), which supports "1 count = 1 second". This is an editorial inference: run one 4-second and one 8-second job and reconcile the invoice before signing.
Why is the real invoice much higher than price × duration?
Four common reasons: retries (a usable clip often takes several generations), separately billed input video or reference images, minimum charges (short clips cost more per second), and 4K/HDR/professional-format surcharges. Budgeting for retries controls the total better than negotiating the unit price.
What does a 30-second spot cost?
At 4 shots × 8 seconds with 3 attempts each (96 seconds generated): Veo 3.1 with audio about USD 38.40, Seedance 2.0 720p about 28.80, Veo 3.1 Fast about 14.40, Gen-4 Turbo about 4.80. Add editing, voice-over, music licensing and review labour.
How much more do 4K and HDR cost?
Per official multipliers: Veo 3.1 4K is USD 0.60/s (1080p with audio is 0.40); Luma HDR is 2× SDR and HDR+EXR 3×; Runway adds 20 credits/s for HDR and 40 credits/s above roughly 4MP. Confirm whether the master really needs these specs.
Is video-to-video more expensive?
Usually. Seedance 2.5 bills output seconds plus input/reference seconds (480p: 20 + 10; 1080p: 68 + 34 credits); Kling 4 with an input video bills the greater of total input duration and output duration. Recalculate for edits and remixes.
Method and limitations
This article uses only official pages accessible on 31 October 2026: Runway Dev's API pricing, Luma's API pricing, and Google Cloud's Vertex AI pricing plus Veo 3.1 documentation. All amounts are converted from official list prices; no enterprise agreements, promo codes or third-party quotes are used. USD is not converted to RMB and no vendor ranking is implied. Runway's credit price, Luma's 5-second block and HDR multipliers, and Google's count definition can change; "count = 1 second" is an editorial inference from two aligned prices, not an official definition, and actual billing follows the console and invoice. Retry rates, the 30-second spot and usable-clip costs are example assumptions to be replaced with your own data. The article does not evaluate output quality, promise results, or constitute copyright or legal advice.
Sources and basis
Data cut-off: 31 October 2026. The following are official pricing/document pages; prices, models and terms follow the original text and invoice.
- Runway, API Pricing & Costs: docs.dev.runwayml.com/guides/pricing/. 1 credit = USD 0.01; per-second credit prices for Veo 3.1, Kling 3.0, Gen-4.5, Gen-4 Turbo, Wan 3.0, Seedance 2.0 and others; input-video, minimum-charge and professional-format surcharges.
- Luma, API Pricing: lumalabs.ai/api/pricing. Ray3.2 per 5-second block, SDR/HDR/EXR multipliers and 5-second list prices.
- Google Cloud, Vertex AI generative AI pricing: cloud.google.com/vertex-ai/generative-ai/pricing. Veo 3.1 / Fast / Lite price per 1 count by resolution.
- Google Cloud, Veo 3.1 generate: cloud.google.com/vertex-ai/generative-ai/docs/models/veo/3-1-generate. 4/6/8-second lengths, up to 4 outputs per prompt.
- OpenAI, Deprecations: developers.openai.com/api/docs/deprecations. Sora 2 / Videos API shut down on 2026-09-24, framing the currently available video API set.