What a 5-Second AI Video Actually Costs
Pricing data · verified 21 Oct 2026Fourteen models, fourteen billing schemes: credits, units, tokens, straight per-second rates. The AI video cost per second is the only figure that makes them comparable, so every number below is stated that way: what a 5-second clip costs when you actually ship it. The cheapest sits 66× below the most expensive.
The same five seconds of video is sold at 66 different prices.
The floor is $0.03, the ceiling is $2.00, and both numbers describe five seconds at the same resolution. What moves between them is not quality. It is who you bought from, and whether that vendor bothered to show a dollar figure before you clicked generate.
The short answer
Most teams are paying 10× more than the work requires
A production-usable 5-second clip costs between $0.03 and $2.00, depending on which model you point at it. Put both ends on the same volume (187,500 billed seconds a month, roughly 500 accepted clips a day once retries are counted) and the monthly bill reads $1,125 against $75,000. Same footage. Same duration. Different invoice.
Billing design accounts for most of that spread, not model quality. Runway sells credits, Kling sells prepaid units, Veo sells tokens through a cloud API. None of the three prints a cost per second next to the generate button, so buyers compare whatever the vendor chose to display, which is rarely the number that matters.
One warning before the table, since most "cheapest AI video" roundups skip it: cost per second and cost per usable clip are not the same number. A model that fails 60% of the time at $0.05 a second is dearer than one that fails 20% of the time at $0.08, and the ranking flips once you count attempts. The acceptance maths is worked through further down.
Table 1
The AI video cost per second table, normalized to one 5-second clip
Sorted cheapest first. All figures are published list rates retrieved 21 October 2026, converted at the vendor's own credit or unit price. Where a vendor sells through both a subscription and an API, the API rate is shown because it is the only one that is directly comparable. Ranges reflect tier differences within one model family.
| Model | Price / sec | Price / 5s clip | Native audio | Note |
|---|---|---|---|---|
| Hunyuan FastTencent · 480p–640p | $0.006 | $0.03 | No | Text-to-video only. No reference images, no negative prompts, low resolution. Cheap for storyboards, not for delivery. |
| MovieFlow 2.0This page720p · 5-second ad model | $0.008 | $0.04 | No | 5-second clips, 10-second render, 720p maximum, silent output. 5 free videos on signup. |
| Hailuo 2.3 FastMiniMax · 768p | $0.032 | $0.16 | No | Billed per clip ($0.19 for a 6s 768p render), so the per-second figure moves with duration. |
| Runway Gen-4 TurboAPI credits | $0.05 | $0.25 | No | 5 credits/sec at $0.01/credit. On the $12/mo Standard plan the effective rate rises to ~$0.10/sec. |
| Veo 3.1 LiteGoogle · 720p | $0.05 | $0.25 | Optional | Cheapest paid tier Google sells. 720p only, no 4K option. |
| Luma Ray 3720p | $0.06 | $0.30 | No | Billed per generation ($0.30 per 5s at 720p). HDR doubles the rate. |
| Kling 3.0Kuaishou · 720p, no audio | $0.084 | $0.42 | No | 0.6 units/sec at $0.14/unit. API access requires a prepaid package starting at $700. |
| Seedance 2.0 MiniByteDance · 720p | $0.095 | $0.48 | Included | Token-billed. Audio generation is free. 4–15s durations, 7 aspect ratios. |
| Pika 2.2720p | $0.10 | $0.50 | Included | Max 5s per generation. Strong on creative effects, weak on character consistency. |
| Veo 3.1 FastGoogle · 720p | $0.10 | $0.50 | Included | Fast tier with synchronized audio. $0.12/sec at 1080p, $0.30 at 4K. |
| Kling 3.01080p, no audio | $0.112 | $0.56 | No | Higher resolution doubles the unit cost. 4K mode jumps to $0.42/sec. |
| Kling 3.0720p, with native audio | $0.126 | $0.63 | Included | Audio adds roughly 50% to the silent rate — worth pricing before you enable it by default. |
| Runway Gen-4subscription, amortized | $0.149 | $0.75 | No | 12 credits/sec against the $28/mo, 2,250-credit Pro plan. Credits expire monthly. |
| Veo 3.1 StandardGoogle · 720p–1080p | $0.40 | $2.00 | Included | Highest fidelity in the category, highest published rate. $0.60/sec at 4K. |
Render times are observed ranges from vendor documentation and independent benchmarks, not contractual SLAs. They shift with queue depth and load. Treat them as order-of-magnitude, not as a promise from any vendor including us.
Table 2
What $100 actually buys
Per-second rates are hard to feel. The same figures are restated below as finished clips, because that is the form a media buyer or a creative director signs off on.
| Model | Clips per $100 | Total footage |
|---|---|---|
| Hunyuan Fast | 3,333 | 4h 38m |
| MovieFlow 2.0 (720p) | 2,500 | 3h 28m |
| Hailuo 2.3 Fast | 632 | 53 min |
| Runway Gen-4 Turbo / Veo 3.1 Lite | 400 | 33 min |
| Luma Ray 3 | 333 | 28 min |
| Kling 3.0 (720p) | 238 | 20 min |
| Seedance 2.0 Mini (720p) | 211 | 18 min |
| Pika 2.2 / Veo 3.1 Fast | 200 | 17 min |
| Kling 3.0 (1080p) | 179 | 15 min |
| Kling 3.0 (720p + audio) | 159 | 13 min |
| Runway Gen-4 (plan) | 134 | 11 min |
| Veo 3.1 Standard | 50 | 4 min |
Monthly column assumes 500 accepted clips per day, 2.5 generation attempts per accepted clip, 30-day month, i.e. 187,500 billed seconds. Same output volume across every row.
Unit conversion
How to turn any vendor's credit system into a cost per second
Three of the fourteen vendors in Table 1 publish a dollar cost per second. The other eleven publish credits, units or tokens and leave the division to the buyer, which is where money gets quietly lost, because the denominator people reach for is usually the wrong one.
One conversion has held on every rate card published this year:
Three ways that formula gets broken.
The first snag is that credits are quoted per call rather than per second, so a 5-second render and a 10-second render carry the same sticker price on the page. The second is monthly expiry: bundled plan credits reset, which turns your real rate into a function of how much of the bundle you burn.
A few vendors also meter compute including queue time, so the identical clip costs more at 9pm than at 3am.
Runway Gen-4 Turbo makes a clean example: 5 credits per second at $0.01 a credit, which is $0.05 a second and $0.25 for a five-second clip.
Buy the $12 Standard plan and the same clip carries a different price, depending on how hard you work the plan. Burn all 2,250 credits and you land near $0.05 a second; burn a third of them and it drifts toward $0.15. The rate card is identical in both cases. Only the utilisation behind it moved.
Which is why every figure here is normalised to one 5-second clip rather than a headline rate. A cost per second that names no resolution, no audio setting and no billing model cannot be compared against anything. It is a number built to be quoted, not used.
Four desks
Four desks, four different invoices
Four buyers spending the same $0.008 a second end up with four very different months. Each card below shows what one of them actually gets, costed from the rates in Table 1. Pick the desk that looks like yours.
01
Paid social · variant testing
The performance marketer testing 400 hooks a month
400 approved variants, 40% acceptance, 5s each. The queue is the bottleneck, not the budget. A 90-second render turns a two-day sprint into a two-week one.
vs $420 on a $0.084/s 720p tier
02
Cross-border DTC · catalogue work
The seller who needs a clip for every SKU
30 products, four creative angles each, 2.5 attempts per accepted clip: 300 generations and 1,500 billed seconds. This is a catalogue job, not an art job, so the per-SKU cost has to stay a rounding error.
vs $48 on a $0.032/s 768p tier
03
Vertical studio · episode production
The studio shipping a 30-episode vertical series
30 episodes × 60 seconds = 1,800 finished seconds, eight attempts per accepted shot. At this volume the rate card decides whether the series is viable at all, and 720p clears the platform floor.
vs $1,210 on a $0.084/s 720p tier
04
Product team · API integration
The team embedding video into their own app
200,000 billed seconds a month behind your own UI. Here the rate is a unit-economics input, not a marketing number, and the line item that decides your gross margin.
$40 per 1,000 clips — a number you can put in a unit-economics sheet
Every figure above is modelled from the Table 1 rates using the assumptions printed on each card. Volume and acceptance rate are the two dials worth turning to match your own operation.
Rerolls and render time
Cost per second is the wrong metric
Text-to-video has a reroll problem. The shot you want rarely arrives on the first generation; it turns up on the third, the fifth, sometimes the eleventh. Which means the price of one generation says very little about the price of a shot you can cut into an edit.
The figure that belongs in a budget is cost per accepted clip: the only per-second number that survives contact with an edit:
Two models at very different sticker prices can land in the same place. A $2.00 clip that you accept 7 times out of 10 costs $2.86 per usable clip. A $0.10 clip that you accept 1 time out of 10 costs $1.00: cheaper per generation, twice as expensive per shot you can actually publish.
Run the direction before arguing for the expensive option, though. For a $2.00 clip to beat a $0.04 clip on short-form work, the cheap model would have to fail about 98% of the time. No two hosted models at 720p differ by anything close to that, so the premium tier wins on capability and almost never on cost per usable clip.
The useful question, then, is how many attempts this model needs before something is shippable, and how long each attempt takes. The second half of that sentence is where budget models tend to fall apart, since it never appears in dollars on anyone's rate card.
Iteration time costs more than generation
A marketer or media buyer costs roughly $35 an hour loaded, and that is the expensive input in this pipeline. Generation sits below it as a rounding error.
Take one shot that needs 20 attempts to get right. At 100 seconds per render (the upper end of what Kling and Seedance typically take for a 5-second clip), that is 33 minutes of somebody watching a progress bar. At 10 seconds per render, the same 20 attempts take 3.3 minutes.
Across those 20 attempts the generation bill difference is only about $7.60, while the labour difference is about $17.50. The waiting costs more than twice as much as the generating.
Scale it to 100 shots a month. The speed gap becomes roughly 50 hours of saved work, about $1,750 at a $35/hour loaded cost. The generation bill, on the 10-second model, would not have reached $80 for all 2,000 renders.
The choice on the table is not cheap model versus expensive model. It is whether somebody on your team spends a week of the year watching progress bars.
Once generation falls below roughly 10% of labour cost, shaving the generation rate stops being optimisation. The variable worth working on is time to a usable shot.
A $0.008/second model that returns in 10 seconds beats a $0.006/second model that takes 90, even though the first is 33% dearer on paper. That comparison is the whole argument.
One caveat: render figures are only comparable when divided by clip length. A vendor quoting "8 seconds" for an 8-second clip is not the same speed as one quoting 10 seconds for a 5-second clip. Convert both to seconds of compute per second of output before you trust the comparison.
The workflow
Three steps, ten seconds
None of this appears on a rate card: the work between having the idea and holding a file you can upload.
Type one line
You type one sentence about the shot. There is no negative prompt to tune, no reference stack to assemble, no settings panel standing between you and a first result.
It renders in 10s
A finished 5-second 720p clip lands before you have switched tabs, which is what turns nine attempts into iteration rather than a queue you stand in.
Keep the one that works
The keeper downloads at $0.04, and so does every reject, which is why binning them never turns into a budget conversation.
The render time quoted on the product page is a figure, not a contractual SLA. It moves with load, and it is better to say so here than have you discover it on a deadline. As for the 720p ceiling: it clears TikTok in-feed and needs one upscale pass for Meta.
FAQ
Frequently asked questions
How do I convert a credit-based pricing page into a cost per second?
Use (price of one credit × credits per generation) ÷ seconds of output. Two mistakes break it: treating credits-per-generation as credits-per-second, and ignoring that bundled plan credits expire monthly.
Runway Gen-4 Turbo is 5 credits per second at $0.01 per credit, which is $0.05 per second and $0.25 for a 5-second clip. On the $12 Standard plan the effective rate depends on how much of the 2,250-credit bundle you burn, which is why plan rows here are labelled amortized.
What is the cheapest AI video model per second in 2026?
On published API rates, Tencent's Hunyuan Fast is cheapest at about $0.006 per second ($0.03 per 5-second clip), but text-to-video only, capped at 480p–640p, no reference images.
The cheapest rate that still supports an ad-ready workflow is around $0.008 per second ($0.04 per clip), capped at 720p. That clears TikTok in-feed and falls short of Meta's 1080×1920.
Need reference images, character consistency or anything above 720p (including from us), and the realistic floor moves to $0.08–$0.10 per second.
How much does a 5-second AI video cost?
Between $0.03 and $2.00 for the same five seconds, a 66× spread for an identical unit of work. Table 1 has all fourteen rows with sources.
These are generation costs. If a shot needs three attempts before you accept it, multiply by three.
How much does it cost to generate 100 five-second videos?
100 clips is 500 seconds of output. At published rates: Hunyuan Fast $3.00 · MovieFlow 2.0 $4.00 · Hailuo 2.3 Fast $15.83 · Runway Gen-4 Turbo $25 · Kling 3.0 720p $42 · Seedance 2.0 Mini $47.50 · Pika 2.2 $50 · Veo 3.1 Standard $200.
Generation only, assuming every clip is usable. At a 40% acceptance rate you generate 250 clips to get 100, so multiply by 2.5.
Why is there such a large price difference between AI video models?
Audio. Models that generate synchronized sound in the same pass (Veo, Seedance, Pika) charge a premium for it. Silent models push that cost to post-production.
Resolution. Kling 3.0 runs from $0.084/sec at 720p to $0.42/sec at 4K, a 5× step for the same duration. Veo 3.1 spans $0.05 to $0.60 across four tiers of one model family.
Business model. Some vendors sell compute at near cost to win developers; others price on output value. A $2.00 clip aimed at a brand film and a $0.04 clip aimed at a social ad are not competing for the same budget line.
Is a cheaper AI video model actually cheaper?
Only if you measure cost per accepted clip. A $0.10 clip you accept 1 time in 10 costs $1.00 per usable shot; a $2.00 clip you accept 7 times in 10 costs $2.86. Neither sticker price predicts the winner.
Add iteration time on top: at a $35/hour loaded cost, 20 attempts at 100 seconds each costs more in waiting than in generation. Once generation drops below ~10% of total cost, render speed dominates the decision.
How long does AI video generation take?
For a 5-second clip, most hosted models return in 30 to 120 seconds. Kling and Seedance sit at the slow end, Runway and Pika at the fast end. Queue depth moves these by 2× or more at peak, and no major vendor publishes a latency SLA on standard capacity.
The fast outlier is a 10-second render, roughly 6–10× the category norm, which is what makes high-volume iteration practical: 20 attempts take minutes instead of half an hour.
Do AI video models generate audio, or do I need a separate tool?
Depends on the model, and it is the largest hidden cost in AI video budgeting. Veo 3.1, Seedance 2.0 and Pika 2.2 generate synchronized audio in the same pass. Runway, Kling (unless you enable audio, +50% on rate), Luma and Hailuo output silent video, so a realistic pipeline adds a voice or music pass plus editing time to sync it.
If the deliverable is dialogue-driven, price the full pipeline rather than the per-second rate.
Can I use the video commercially?
Generally yes on paid tiers; free tiers usually exclude it. Runway grants commercial rights on all plans. Kling, Seedance and Luma include commercial licences on paid plans with no attribution. Free tiers commonly add watermarks and restrict commercial use.
Verify two things before publishing client work: whether the licence covers the specific use (paid ads, broadcast, resale), and whether your input assets are cleared. Vendor terms do not transfer rights you never had in the source material.
Should I pay per clip or buy a subscription for AI video?
Start pay-as-you-go. Stay on API or per-clip billing until you are consistently spending more than $25–$30 a month with one provider, then move to that provider's subscription.
Two traps sit under that advice. Plan credits expire monthly, so an under-used plan carries a far higher effective rate than its sticker: the $12 Runway Standard tier works out near $0.10 a second for Gen-4 Turbo, double the API rate. Prepaid API packages also come with expiry dates rather than discounts: Kling's start at $700.
Which AI video model should I use for TikTok or Reels ads?
Short-form performance advertising pays for volume and speed ahead of fidelity. You are testing dozens of hooks, not grading a brand film, which points at the $0.008–$0.10 band with fast renders and vertical output.
One spec check: TikTok accepts in-feed down to 540×960, so 720p vertical clears it. Meta Reels asks 1080×1920, so a 720p model needs an upscale there. If most spend is TikTok the ceiling rarely bites; if Meta, price the upscale in.
The workflow that holds up: generate every variant on a cheap, fast model, find the one or two that keep attention in the first two seconds, then re-generate only those through a premium model, and only for the placements you are paying to amplify.
How often is this AI video pricing comparison updated?
AI video pricing changes monthly. Google restructured Veo into four tiers, Kling moved to a unit system, and Sora 2's API reached a published shutdown date, all within 2026. Last verified against vendor pricing pages on 21 October 2026.
Figures are published list rates, not negotiated enterprise pricing, and exclude taxes. Where vendor documentation conflicted with third-party summaries, the vendor figure was used.
Methodology
How these numbers were produced
Every rate on this page was taken from the vendor's own published pricing or developer documentation on 21 October 2026, then converted to dollars per billed second of output. Credit-based vendors were converted using the vendor's own stated credit price, not a reseller's. Where a model is sold at multiple tiers, the tier is named in the row rather than averaged away.
Enterprise and negotiated pricing, reseller and aggregator rates, and subscription plan prices are excluded from the headline per-second figures. The one exception is a row explicitly labelled as amortized, because plan amortization depends on utilization and is not comparable to an API rate.
Render times are observed ranges from vendor documentation and independent benchmarks. No vendor in this table, including us, publishes a contractual latency guarantee on standard capacity. Treat them as directional.
Sources
Each figure links to the page it was read from, so you can verify any row yourself.
- Google · Veo 3.1 Lite, Fast, StandardGemini API pricing
- Google Cloud · Veo on Vertex AIVertex AI pricing
- Runway · Gen-4, Gen-4 Turborunway.com/pricing · API credit docs
- Kuaishou · Kling 3.0Kling API documentation
- Luma · Ray 3Luma API pricing
- Pika · Pika 2.2pika.art/pricing
- MiniMax · Hailuo 2.3 FastMiniMax pay-as-you-go pricing
- ByteDance · Seedance 2.0Volcengine Ark model pricing
- OpenAI · Sora 2OpenAI API pricing
MovieFlow figures are drawn from our published plan and credit pricing. All rates were read on 21 October 2026. Vendor cards change without notice; confirm on the linked page before committing budget.
If you are checking how credits convert into billable seconds on our own product rather than on a competitor's, the credits and pricing chapter of the tutorial walks through the same arithmetic on a long film and a short clip.
For the questions people ask before subscribing (what the free tier covers, how credits are spent, where the price list lives), the product FAQ answers them directly.
Try it at $0.008 per second
One line of text in, one 5-second video out, in about 10 seconds. Sign up and your first five videos are free.
Subscribe — see live plans →5 free 5-second videos on signup · $0.008/sec is our standing rate, not a promotional price
Plans, priced the same way as the table above
Annual billing shown, because that is the rate most people actually pay. The struck-through figure is the monthly price, and the SAVE badge is MovieFlow's own discount. Every plan includes the same model set: you are buying output volume and concurrency, not features. Monthly billing is available on the plan page.
- ✓1080P HD, watermark-free downloads
- ✓50GB storage · 1 concurrent job
- —No peak-time protection
- ✓Everything in Standard
- ✓200GB storage · 5 concurrent jobs
- ✓Peak-time protection · priority support
- ✓Everything in Pro
- ✓1TB storage · 10 concurrent jobs
- ✓Highest monthly output allotment
What each plan actually buys. Figures above are the clips you can render per month on the 720p tier this page prices, at 5 seconds each. Higher resolutions and heavier models render fewer clips for the same plan. Drop to a smaller output or a higher tier if your work needs 1080p or 4K.
Annual billing is shown because it is the rate most customers pay; the struck-through price is monthly. Prices are in US dollars and exclude tax.