Your vision.
In perspective.
An AI video generation cost calculator for the models creators actually use, from Veo and Runway to Kling and Luma. One clear view of what your next video will cost, from the first generation to the final cut.
8 s × 3.5 attempts × 30 clips × $0.12/s
Google · Veo 3.1 Fast · 1080p$0.96 per 8 s clip and $7.20 per minute, before attempts
Google (Veo 3.1 / Gemini Omni Flash). Veo 3.1 rates are quoted as 'video with audio price (default)' on Google's pricing page. Full breakdown · Vendor pricing page ↗
Compare model costs
Lowest verified tier per model.
| # | Model / tier | Base rate | Usable clip | Monthly | Source |
|---|---|---|---|---|---|
| 01 |  Hailuo / MiniMax (H3 / Hailuo 2.3)LegacyHailuo-02 legacy, 10s clip · 512p | $0.015per second | $0.42with 3.5× attempts | $12.60 | Source ↗ |
| 02 |  Grok Imagine (xAI, Video 1.5 / 1.5 Lite)Grok Imagine Video 1.5 Lite · 480p | $0.02per second | $0.56with 3.5× attempts | $16.80 | Source ↗ |
| 03 |  Wan (Alibaba, 3.0 / 2.7)Wan 2.6 Flash image-to-video, silent · 720p | $0.025per second | $0.70with 3.5× attempts | $21.00 | Source ↗ |
| 04 |  Luma Dream Machine (Ray3.2 API)T2V/I2V, 5s clip · 540p | $0.03per second | $0.84with 3.5× attempts | $25.20 | Source ↗ |
| 05 |  LTX (Lightricks)LTX-2.3 Fast · 720p | $0.03per second | $0.84with 3.5× attempts | $25.20 | Source ↗ |
| 06 |  Seedance (ByteDance, 2.5 / 2.0)Seedance 2.0 Mini, 5s clip · 480p | $0.036per second | $1.01with 3.5× attempts | $30.24 | Source ↗ |
| 07 |  Pika (2.5 API)Pika 2.5 · 720p | $0.04per second | $1.12with 3.5× attempts | $33.60 | Source ↗ |
| 08 |  Google (Veo 3.1 / Gemini Omni Flash)Veo 3.1 Lite · 720p | $0.05per second | $1.40with 3.5× attempts | $42.00 | Source ↗ |
| 09 |  Runway (Gen-4.5 / Gen-4 Turbo / Aleph 2.0)Gen-4 Turbo · 720p/1080p | $0.05per second | $1.40with 3.5× attempts | $42.00 | Source ↗ |
| 10 |  HappyHorse (Alibaba, 1.1 / 1.0)HappyHorse 1.1 · 480p | $0.07per second | $1.96with 3.5× attempts | $58.80 | Source ↗ |
| 11 |  Kling AI (3.0 / 3.0 Turbo / Omni)Kling 3.0 no audio · 720p | $0.084per second | $2.35with 3.5× attempts | $70.56 | Source ↗ |
Different tiers and resolutions are shown. This is a cost comparison, not a ranking of video quality. Credits and estimated rates may differ from direct API billing.
OpenAI Sora API shut down Sep 24, 2026 — see /sora-api-shutdown/
How the estimate works
Rate × clip length × attempts × monthly clips. Change any input to recalculate.
Keep the assumptions visible
Attempts are your planning assumption. Actual generation success varies by shot and model.
About these rates
Rates are list prices read from each vendor's own pricing or documentation page and last checked Oct 20, 2026. They are estimates, not quotes.
Quick answer
For a straight per-second sticker price, Hailuo / MiniMax (H3 / Hailuo 2.3) is the cheapest tracked tool right now at roughly $0.015/sec (Hailuo-02 legacy, 10s clip, 512p). That is a legacy model; the cheapest current-generation tier is Grok Imagine (xAI, Video 1.5 / 1.5 Lite) at $0.02/sec (Grok Imagine Video 1.5 Lite, 480p). The sticker price is not the real price, though: practitioner reports (cited below) put the number of generations per usable clip at 3-6 or more once you account for character drift, wrong motion and artifacts, so the calculator above defaults to a 3.5x retry multiplier on top of the base rate, adjustable from 1x to 6x. Audio is priced differently too: Google Veo 3.1 rates are quoted with audio included, Kling lists native-audio and no-audio rates separately, and Runway prices audio models separately from video.
One price is not comparable across vendors without normalizing first: Kling quotes in units, Runway in credits, MiniMax prices its older Hailuo models per clip, and both Luma and Pika change the per-second rate with clip length. Pick your tool, resolution and volume in the calculator above to get one number.
Notes on each vendor’s pricing
Google (Veo 3.1 / Gemini Omni Flash)
Token-billed: 5,792 output tokens per second of 720p video at $17.50 per 1M tokens; Google states this is approximately $0.10 per second.
Audio: Veo 3.1 rates are quoted as 'video with audio price (default)' on Google's pricing page.
Runway (Gen-4.5 / Gen-4 Turbo / Aleph 2.0)
Runway API credits cost $0.01 each. Runway's developer pricing page also resells third-party models (Kling, Veo, Seedance and others); those are not listed here.
56 credit ($0.56) minimum per generation.
Audio: Video generation is priced separately from audio models (for example Seed Audio at 0.25 credits per second, 5 credit minimum).
Kling AI (3.0 / 3.0 Turbo / Omni)
Kling quotes API prices in units; 1 unit = $0.14 at list price. Figures here are the dollar amounts printed beside each unit price.
Audio: Kling lists no-audio and native-audio rates separately.
Luma Dream Machine (Ray3.2 API)
Per-second rates are not linear in clip length. Prices shown are SDR output; HDR is 2x and HDR + EXR is 3x.
Audio: Luma's API pricing page does not state whether output includes audio; the rates listed are for video output, so budget audio separately.
Hailuo / MiniMax (H3 / Hailuo 2.3)
Hailuo 2.3 and Hailuo-02 are listed under 'Legacy Models' and priced per clip; per-second figures for those are clip price divided by clip length. MiniMax H3 is priced per second.
Audio: MiniMax's pricing and guide pages do not state whether output includes audio.
LTX (Lightricks)
Text-to-video and image-to-video rates are identical per tier. LTX-2.5 is priced above LTX-2.3 at every resolution.
Audio: LTX's API documentation (docs.ltx.video) states that all endpoints return video with synchronized audio.
Pika (2.5 API)
Rates are from the Pika API price list (dev.pika.art), which also lists a $10/month API Club membership fee. The 1080p rate depends on clip length: 5s clips are listed at a higher per-second rate than the base 1080p rate. The consumer plans on pika.art are credit-based across many models and are not priced here.
Listed as the base 1080p rate; the 5s-clip rate is a separate override and 5 and 10 seconds are the only durations the API accepts.
Audio: Pika's Pika 2.5 page states that renders come back silent.
Wan (Alibaba, 3.0 / 2.7)
International (Singapore) deployment rates in USD from Alibaba Cloud Model Studio. Other regions are priced differently.
List price. Alibaba Cloud's page advertises a limited-time 30% discount beside this list price; the discounted amount is not stated, so the billed rate may be lower while the promotion runs.
Audio: Alibaba lists Wan 2.5 and later as audio-video output; Wan 2.6 Flash also has a cheaper silent option.
Seedance (ByteDance, 2.5 / 2.0)
Sold through BytePlus ModelArk and billed per token (USD per million tokens). Per-clip prices are BytePlus's own published examples for a 5-second 16:9 clip with no video input; tokens scale with resolution and output length, so longer clips cost proportionally more (2.0 series: 4 to 15 seconds; 2.5: up to 30 seconds). Inputs that include video are billed at a different token rate and cost more (for example Seedance 2.5 at 720p is listed at $1.244 to $4.838 per 5-second clip depending on input-video length). Seedance 1.5 Pro and 1.0 Pro are also sold and are not listed here.
Audio: BytePlus lists synchronized audio and video generation for Seedance 2.5 and a generate_audio option for the 2.0 series; the price list has one rate per resolution with no separate audio line.
Grok Imagine (xAI, Video 1.5 / 1.5 Lite)
Billed per second of generated video; clips run 1 to 15 seconds (default 8). Image inputs are billed in addition: $0.01 per image on Video 1.5 and 1.5 Lite, and on Grok Imagine Video $0.002 per image plus $0.01 per second of input video.
xAI states Lite reaches 1080p for text-to-video and image-to-video by upscaling 720p.
Audio: xAI states generated videos include an audio track by default (lip-synced speech on Video 1.5 and 1.5 Lite) and that generate_audio=false returns a silent video; the price list has one per-second rate per resolution.
HappyHorse (Alibaba, 1.1 / 1.0)
HappyHorse is a model family sold on Alibaba Cloud Model Studio. International (Singapore) deployment rates in USD for text-to-video; image-to-video and reference-to-video are priced identically per model. Clips run 3 to 15 seconds at 24 fps. The video-editing model (happyhorse-1.0-video-edit) is priced separately and not listed here. Other regions are priced differently.
Audio: Alibaba Cloud Model Studio lists Audio as a feature of every HappyHorse model and describes text-to-video as generating videos with audio; there is no separate audio rate.
The retry multiplier
No vendor publishes a “clips per keeper” number, so the default 3.5x is a stated assumption, not a fact. Adjust the slider to your own workflow. Two practitioner sources back the 3-6x range it defaults from.
- InstrumentalFx - Why AI Video Generation Takes More Attempts Than It Should“3-6 failed generations per keeper is typical; the industry-observed range runs 3-10 attempts per usable shot.”
- TopVidTools - What AI Video REALLY Costs: Yield Rates & True Price Per Usable Clip (2026)“Generation-to-final ratio commonly runs 5:1 or higher, sometimes 10:1 for shots requiring precise continuity.”
FAQ
How much does AI video generation cost?
It depends entirely on tool, resolution and how many retries it takes to get a usable clip. Base API rates as of Oct 20, 2026 range from about $0.015/sec (Hailuo / MiniMax (H3 / Hailuo 2.3), Hailuo-02 legacy, 10s clip, 512p) to $0.778/sec (Seedance (ByteDance, 2.5 / 2.0), Seedance 2.0, 5s clip, 4K). The cheapest tracked tool today is Hailuo / MiniMax (H3 / Hailuo 2.3) at roughly $0.015/sec to $0.13/sec depending on tier, and the true cost per usable clip is higher once you account for retries - use the calculator above with your own retry assumption.
How many videos can I make with a given credit or dollar budget?
Divide your budget by the tool's true cost per usable clip (base price/sec × clip length × your retry multiplier), not by the sticker per-second rate. At $0.06/sec with 8-second clips and no retries, a $50 budget buys about 104 clips on paper - but at a realistic 3.5x retry multiplier for character-consistent shots, that same $50 buys closer to 30 usable clips. The calculator above does this division live for whichever tool, resolution and retry assumption you pick.
Why do AI video costs vary so much across sources?
Three reasons: vendors price by resolution, clip length and model generation differently, and they handle audio differently (Veo rates are quoted with audio, Kling lists native-audio and no-audio rates separately, Runway prices audio models apart from video); they quote in different units (per second, per clip, credits, Kling units); and per-second sticker prices ignore the retry rate needed to get a usable clip - which practitioner reports put at 3-10 attempts depending on the shot. This page normalizes all of that into one $/sec and true-cost-per-clip figure so you can compare tools on the same basis.
What is a 'true cost per usable clip' and why isn't it just the sticker price?
The sticker price is what one generation costs. The true cost is what a keeper costs - the clip you actually publish. Character drift, wrong camera motion and artifacts mean most workflows need several generations per keeper. The calculator multiplies the sticker price by a retry multiplier (default 3.5x, adjustable 1-6x on the slider above) to estimate that real number - see the retry multiplier section above for the practitioner sources behind that default.
Is Kling AI's pricing on this page official?
Yes. The Kling figures are the per-second API list prices Kling publishes at kling.ai/dev/pricing, quoted there in units and converted to dollars here. They are list prices; Kling's contact form for larger API usage notes that pricing depends on models, features and scale, so a negotiated rate can differ. Confirm the rate in your own Kling developer account before committing budget.
What happened to OpenAI Sora's API?
It has shut down. OpenAI's help center says the Sora web and app experiences were discontinued on April 26, 2026, and its deprecations page lists the Videos API and the sora-2 and sora-2-pro models as shut down on September 24, 2026. Sora is no longer in the calculator; see /sora-api-shutdown/ for the full timeline and alternatives.
Sources
See also
- AI coding cost calculatorSeat and usage costs for coding assistants
- Stack savings calculatorWhat a leaner software stack saves
- Sora API shutdownTimeline and alternatives
- Is my app dying?Shutdown tracker for apps and tools
- App swapFind an alternative to any app
- AI toolsDirectory of AI tools
Rates from official vendor pages, checked Oct 20, 2026. Estimates, not quotes — confirm before committing budget.