How much does AI video generation cost in 2026? Per-second prices, per-shot reality

Cost table across the twelve keys

Twelve keys lotgen.ai pastes: Google, BytePlus ARK, fal, evolink, Kling, OpenAI, MiniMax, Black Forest Labs, Alibaba, ElevenLabs, xAI and Runway. Luma and Sora are models on fal or evolink, not keys. BytePlus ARK is the console that issues the Seedance key, and the name to put on the invoice.

Per-second list from official pages, 2026-09-18. One comparable row per key. Audio-on where that key publishes a split. 720p where that key publishes 720p. Marketplace rows are labeled. Missing cells are omitted, not guessed.

How to read this table

• Dollars and seconds only. 3 attempts is a modeling assumption, not a measured keep rate. Multiply the one-clip price by however many takes you actually run. There is no lotgen.ai observed retry rate in this page.
• Veo 3.1 on the Gemini / Vertex predict path does not generate 10 seconds in one take. An 8-second row is the closest fair compare with Kling, Runway, xAI, BFL, Alibaba, and the Kling-shaped marketplace rows. A 10-second Kling row is in the 1080p table below.
• fal and evolink are their own checkout. Matching Kling’s 720p audio row on fal on the day checked is a coincidence, not a promise that 1080p or 4K match.
• BytePlus per-second is computed: official 5.6 USD / M tokens, official formula (input seconds + output seconds) × width × height × frame rate / 1024, 24 fps, 1280×720, no input video, plus the one closing frame ARK bills (lotgen measured it on five real jobs; the page’s own examples reproduce without it, so these cells sit a cent or two above the page). A 10-second take is 216,900 tokens, $1.21. Actual bill is usage.completion_tokens. The same page prints a time-limited 25% off on 480p/720p Seedance 2.0 Fast through 2026-10-07 14:00 UTC+8 (about $0.09/s at 720p). This table uses the list 5.6 rate. That window is dated copy, not a deadline on this page.
• Alibaba dollars are the International wan2.7-t2v row on Model Studio ($0.10/s 720P, $0.15/s 1080P). Do not paste evolink Wan 3.0 (~$0.038/s at 480p) as Alibaba list. The product catalog on live /pricing still prints Wan 2.7 1080p as unpriced; that is lotgen’s adapter catalog, not Alibaba’s page.
• MiniMax dollars are official PAYG list. The video-package page still bills Hailuo 2.3 in points (768p/6s = 1 pt, 768p/10s = 2 pt, 1080p/6s = 2 pt). A 5% Standard package ($1,000 / 3,760 points) still infers about $0.28/point. That inference matches PAYG 768p 6s ($0.28) and 768p 10s ($0.56). It does not match PAYG 1080p 6s ($0.49). Use PAYG as list.

1080p, audio on, where the official page prints 1080p

Sibling compare (live): Kling vs Veo cost uses the 8-second row because that is the length Veo 3.1 will bill. Sibling Kling credits (live): Kling AI pricing explained owns the app-credit walkthrough. The Veo-only walkthrough is Veo 3 cost per second. This pillar uses the public API meters.

Three pricing models: credits, per-second API, seat plus usage

Almost every “how much does AI video cost” disagreement is a mixed meter.

1. Subscription credits (app plans)

You buy a monthly grant. The generate chip tells you how many credits this take burns. Unused grant often expires. Kling’s website app is this meter: Standard 660 credits list $10.00, Pro 3,000 credits list $37.00, Premier 8,000 credits list $92.00, Ultra 26,000 credits list $180.00, from the live Kling credits walkthrough. Official Credits Policy peg: $1 USD = 66 credits. Membership credits last about a month. Purchased credits last 2 years. There is still no public credits-per-second table for the app; the chip is the source of truth.

MiniMax video packages are the same shape with a different unit: Standard $1,000 for 3,760 video points (page labels SAVE 5%), 1-month validity. Hailuo 2.3 burns 1 point for 768p/6s and 2 points for 768p/10s or 1080p/6s. PAYG is a separate official dollar table (see the MiniMax rows above).

Pro: predictable monthly ceiling if you actually use the grant. Con: expiry turns unused balance into a higher effective dollar per kept shot. A cheap per-credit plan you do not burn is more expensive than a dearer pay-per-success API.

2. Pay-per-second (or per-token) API

The lab publishes a dollar row. You pay if the take succeeds. Kling developer API: 1 video Unit = $0.14, per second, failed video generations do not deduct. Google Gemini Veo: per second, charged only if the video is successfully generated. Black Forest Labs: 1 credit = $0.01, per second of output. Runway Dev: credits at $0.01, gen4.5 at 12 credits/s. xAI: per second by resolution. BytePlus ARK: tokens from pixel area, then USD per million tokens. Alibaba Model Studio Wan 2.7 t2v: per second by resolution on the International row. MiniMax PAYG Hailuo: per take at a printed dollar, not per second.

Pro: comparable across labs if you hold resolution, audio, and billed seconds fixed. Con: “per second” hides the length the API will accept (Veo 3.1: 4, 6, 8) and the row you actually land on (Fast vs Standard, silent vs audio).

3. Seat plus usage (control plane on your keys)

lotgen.ai charges a seat. You still pay the lab for seconds, at list, on your own keys. Public seat copy on lotgen.ai/pricing on 2026-09-18: $99 Solo, Team $90 a seat from 3, $79 from 6. The product does not resell Kling credits or Veo tokens.

Pro: the quote is in dollars before the take runs, across the keys you actually paste. Con: it is a second line item. Budget seats and seconds separately.

A deeper credits-versus-pay-as-you-go walkthrough is coming; until then, this section is the summary.

fal and evolink sit next to the lab APIs as marketplace keys. Their catalog rate is their invoice, not Google’s or Kling’s. On 2026-09-18, fal’s Kling 3.0 Standard 720p audio-on row is $0.126/s, which matches Kling official. evolink’s Kling 3.0 page lists $0.08/s base (720p audio on = $0.12/s) and tells readers official Kling is $0.089/s. Kling’s own page that day is $0.084/s silent / $0.126/s audio at 720p. evolink’s “official” figure is wrong. Use it only as proof that published numbers disagree.

Cost per usable shot: the retry multiplier

List price per second is the sticker. The number that hits the card is:

usable-shot cost = list $/s × billed seconds × attempts

Three things move that number more than the brand name.

1. Attempts. A rejected motion, a bad hand, a missed product label: you pay for the miss and the keep. This page models 3× in the tables. Change the multiplier to your own keep rate. No published “industry retry rate” is cited here because the public pages checked for this page do not give one. 3× is not a benchmark.

2. Billed seconds, not requested seconds. Ask Veo 3.1 for 10 seconds and you do not get a 10-second invoice. You get a refusal, then you generate 8 seconds (or two clips). Vertex generate-video docs on 2026-09-18: Veo 3 models accept 4, 6, 8 seconds, default 8. MiniMax Hailuo 2.3 at 1080p exists only at 6 seconds. Kling 3.0 in lotgen’s adapter: 3 to 15 seconds (lotgen’s own test, 2026-08-25, not Kling marketing copy).

3. The row, not the family. “Veo is $0.40 a second” is Standard with audio on Gemini. Vertex silent Standard is $0.20/s. Fast 720p with audio is $0.10/s. Lite 720p is $0.05/s. fal’s Veo 3.1 Fast with audio is $0.15/s, which disagrees with Google Fast $0.10 / $0.12. That is a reseller row, not a Google typo.

Worked 8-second, 720p, audio-on (or default audio) usable shot at 3 attempts, official or labeled marketplace, 2026-09-18:

Lite Veo is cheaper than Kling at this length. Standard Veo is about 3.2× Kling 3.0 720p audio ($3.20 vs $1.01). Fast Veo 720p is cheaper than Kling 3.0 with audio ($0.80 vs $1.01). Alibaba Wan 2.7 t2v International 720P matches Fast Veo at $0.80 for an 8-second take. “Which is cheaper” is the tier, the billed length, and the keep rate, not the logo.

The live calculator at lotgen.ai/tools/ai-video-cost-calculator lets you set the attempt slider. The 3-attempt slider stays a model you set, not a measured keep rate.

What drives price within one provider

Hold the brand still. These four knobs still move the bill.

Resolution. Kling 3.0 native audio: 720p $0.126/s, 1080p $0.168/s, 4K $0.42/s. Veo 3.1 Standard is $0.40/s at 720p and 1080p, then $0.60/s at 4K. Veo Fast is the row that actually splits 720p ($0.10) from 1080p ($0.12). xAI 1.0 publishes 480p $0.05 / 720p $0.07 and no 1080p cell. xAI 1.5 publishes 480p $0.08 / 720p $0.14 / 1080p $0.25. BFL t2v: hd $0.17, fhd $0.29, qhd $0.40, uhd $0.80. Alibaba Wan 2.7 t2v International: 720P $0.10/s, 1080P $0.15/s. BytePlus tokens scale with width × height, so 1080p is not a small step up from 720p (Seedance 2.5 list: about $0.232/s at 720p vs about $0.571/s at 1080p, same formula and a higher token rate at 1080p).

Duration. A 10-second clip is not 25% more than an 8-second clip if the API will not accept 10. Veo 3.1 predict: 4, 6, 8. Gemini Omni Flash is a different Google video model: the generate-video docs allow a duration string from 3s to 10s, and the Gemini pricing page prices output video at $17.50 / 1M tokens, 5,792 tokens per second of 720p, “approximately $0.10 per second.” Do not copy Omni lengths onto a Veo cost table. MiniMax 1080p is 6s only.

Audio. Kling 3.0 720p/1080p native audio is a separate rate row (about +50%), not a multiplier you apply after the fact. 4K is flat $0.42/s either way. Gemini Veo 3.1 lists with-audio as the default and does not print a silent row. Vertex prints both: Standard 720p/1080p silent $0.20, with audio $0.40. If one blog says Veo is $0.20/s and another says $0.40/s, they may both be quoting Google. BFL’s published video table does not split audio. BytePlus Seedance has no published audio surcharge. Alibaba’s International Wan 2.7 t2v row does not print a silent/audio split on the Model Studio table fetched on the day checked.

Fast vs standard (and Turbo, Lite, Pro). Veo Fast 720p is one quarter of Veo Standard ($0.10 vs $0.40). Kling 3.0 Turbo with native audio is 0.8 Units/s at 720p ($0.112) and 1.0 Unit/s at 1080p ($0.14). MiniMax Hailuo 2.3 Fast PAYG is $0.19 / $0.32 / $0.33 against Hailuo 2.3 $0.28 / $0.56 / $0.49 at 768p 6s / 768p 10s / 1080p 6s. Quality is not in these tables. A cheaper second is not a better take.

The Veo-only walkthrough (Standard / Fast / Lite, Gemini vs Vertex, silent vs audio, the 10-second trap) is Veo 3 cost per second. The Google rows on this pillar and the sibling Kling vs Veo cost are the same dated Veo numbers.

Budgeting by use case

Assumptions for the three sketches: official 2026-09-18 rows, 3 attempts per kept shot unless noted. These are budget sketches, not invoices. Formulas are in the previous section so you can swap the multiplier.

1. Ad variant testing (many prompts, keep-rate matters)

You need 20 kept 8-second 720p shots this week, audio on / default.

If the test is “did the product read” more than “did it look like a film,” Lite or Fast is the cheaper official Google row. If you already hold Kling Units that will expire with the resource pack, burn Kling on the volume tests first. Unused Kling API balance does not carry over.

A 50-variant ad test is its own page, AI UGC ads cost, not this pillar. This page only needs the per-shot arithmetic.

Request an invitationSeats and pricing

2. Client deliverables (1080p, audio, fewer throws)

One hero for a client, 1080p, audio on, you budget 3 takes and deliver one. Veo billed length 8 seconds. Kling billed length can be 8 or 10; the 8-second row is the fair compare.

If the client contracted “Veo 3.1” without a tier, get the tier in writing. The gap between Fast and Standard at 8 seconds is $0.96 vs $3.20 per take.

Request an invitationSeats and pricing

3. Solo creator iteration (you will throw most of it away)

50 exploratory takes, 720p, you keep a handful. Count is takes, not kept shots, so the attempt multiplier is 1×.

If you iterate silent and add audio later, Kling’s silent 720p row ($0.084/s) and Vertex’s silent Standard row ($0.20/s) are the comparable pair, not Gemini’s $0.40/s default. If you iterate with audio because that is what you ship, use the with-audio rows or you will under-budget by 2× on Vertex Standard.

Monthly ranges from the same formula, labeled estimates: 20 kept 8s 720p shots per month at 3 attempts is the “20 kept” column above. 80 kept shots is 4× that column. Swap the attempt slider before you treat any of those as a forecast.

Request an invitationSeats and pricing

How to see the price before you spend

On lotgen.ai the same configuration is meant to appear as a dollar quote before the job runs, on your own keys, across the twelve platforms the product actually keys. Luma and Sora stay models on fal/evolink. BytePlus ARK is the console that issues the Seedance key, and the name to put on the invoice. You still pay the provider. lotgen does not resell their credits.

Invite-only beta. Request access: lotgen.ai/request. Seat and plan detail: lotgen.ai/pricing. Brand, agency, creator, and developer are the four tiles on that form; pick the one that matches. Developers using API keys can use the same form: request an invitation.

A closer look at gateway versus lab invoices, and a quality compare, are coming as pages of their own. This one is the cost pillar.

Request an invitationSeats and pricing

Calculator

The calculator behind these pages is dated the same way: length, resolution, audio, usable shots, attempts, the twelve keys.

Live path: lotgen.ai/tools/ai-video-cost-calculator. Its defaults: 10 seconds, 1080p, audio on, 10 usable shots, 3 attempts, all twelve keys. Veo will clamp to 8 billed seconds in the row. The 3-attempt slider stays a model you set, not a measured keep rate. The door from it is the same invitation form, and the tile you pick there (brand, agency, creator, developer) is how the answer gets worded for you.

The calculator and this page read one dated rate table, so the Alibaba and MiniMax rows that became official list on 2026-09-18 moved on both at once.

Recompute from the tables above if you do not want the tool:

billed_seconds = clamp(requested_seconds, provider min, provider max)
cost_per_clip  = list_rate × billed_seconds
cost_per_usable_shot = cost_per_clip × attempts
batch_total    = cost_per_usable_shot × usable_shots

Prices checked on 2026-09-18: how to re-check yourself

Refresh this page monthly, and within a few days of a major model or plan change. Full rewrite when a lab replaces the default video model.

1. Kling Units and per-second rows: kling.ai/dev/pricing. Confirm 1 Unit = $0.14 for video, then the Kling 3.0 720p/1080p/4K cells with and without native audio. App credits live on Kling AI pricing explained.
2. Veo 3.1 Gemini per-second rows: Gemini API pricing, section “Veo 3.1”. Confirm Standard / Fast / Lite and 720p vs 1080p vs 4K.
3. Veo silent vs audio, and Veo 3 lengths: Vertex generative AI pricing and the generate-video docs (Veo 3: 4, 6, 8 seconds, default 8).
4. BytePlus ARK: ModelArk pricing. Confirm Seedance 2.0 Fast 5.6 USD/M tokens (list), whether the 25% off window is still printed, and 1080p support. Recompute tokens with the published formula. Do not paste a remembered 4.2 promo rate as list. Seedance 2.5 1080p list is 11.7; the 28% off window printed through 2026-09-17 14:00 UTC+8.
5. fal Kling: fal Kling Video v3 Standard and Pro. Marketplace. Do not paste fal Fast Veo $0.15/s over Google Fast $0.10/$0.12.
6. evolink Kling: evolink.ai/kling-3 and evolink.ai/pricing. Marketplace. Ignore their “official Kling $0.089” line unless Kling’s own page has moved to that number.
7. MiniMax: Video packages for the point grid, and PAYG for Hailuo dollars. Use PAYG as list. The 5% package figure is an inference, not a price.
8. Black Forest Labs: docs.bfl.ml/quick_start/pricing. Confirm 1 credit = $0.01 and the t2v hd/fhd/qhd/uhd row.
9. Runway: Dev pricing. Confirm gen4.5 at 12 credits/s and $0.01/credit.
10. xAI: grok-imagine-video (480p $0.05, 720p $0.07) and grok-imagine-video-1.5 (480p $0.08, 720p $0.14, 1080p $0.25).
11. OpenAI Sora (model on fal/evolink, not the OpenAI key): OpenAI API pricing. Sora 2 $0.10/s at 720p, Sora 2 Pro $0.30/s at 720p. Batch is half on that page.
12. Alibaba: Model Studio pricing. International wan2.7-t2v is $0.10/s 720P and $0.15/s 1080P. Do not paste evolink Wan 3.0 as Alibaba official.
13. ElevenLabs: elevenlabs.io/pricing. Voice plans only, unless they publish a video list price.

Recompute: $/s × billed seconds × attempts. If the billed seconds are not a length that provider accepts, stop and check the model id.

Changelog of this page

Related on this site: Kling vs Veo cost · Kling AI pricing explained · Veo 3 cost per second · AI UGC ads cost · AI video cost calculator · how a render is quoted before it runs · seats and pricing · request an invitation

Request an invitationSeats and pricingSee a dollar quote on your prompt, on your own keys, before it runs. Team walkthroughs are 20 minutes and sit in the Team aside above.

FAQ