AI video cost calculator
Enter length, resolution, audio, how many usable shots you need, and how many attempts you usually take. The table uses each key’s official list price, dated, and turns that into dollars per clip, per usable shot, and for the batch.
Prices checked on 2026-09-18 against each provider’s own pricing page. Third-party and marketplace rows are labeled. A missing cell is never a guess: it is marked unverified or omitted.
Request an invitationSeats and pricing
Your spec
What this spec costs
One clip is one render of 10 seconds at 1080p. The batch is 10 usable shots × 3 attempts = 30 renders.
| Key | Model billed | List price | Billed seconds | Cost per clip | Per usable shot | Total for this batch | Notes |
|---|---|---|---|---|---|---|---|
| MiniMax | Hailuo 2.3 | $0.49 / take | 6 billed (asked 10) | $0.49 | $1.47 | $14.70 | clampedOfficial: MiniMax's pay-as-you-go page prints Hailuo 2.3 in dollars per take.The video-package page bills the same take at 2 points; a 5% package ($1,000 / 3,760) infers $0.28 a point, which comes to $0.56, not this cell.1080p is offered only at 6 seconds.Audio ignored: the point grid does not split it.6 s billed (asked 10). Hailuo 2.3 at 1080p bills 6 seconds. |
| Veo 3.1 Fast | $0.12/s | 8 billed (asked 10) | $0.96 | $2.88 | $28.80 | clampedOfficial Gemini rate, audio on by default.8 s billed (asked 10). Veo 3.1 takes 4, 6 or 8 seconds in one take. | |
| Runway | gen4.5 | $0.12/s | 10 · 720p delivered | $1.20 | $3.60 | $36.00 | clampedOfficial: 12 credits/s × $0.01, one flat rate with no resolution column.Delivered at 720p: gen4.5 text-to-video accepts 1280:720 and 720:1280 and no larger pair (lotgen’s own test, 2026-09-06).Audio ignored: the published rate does not split it. |
| Alibaba | Wan 2.7 t2v (International) | $0.15/s | 10 | $1.50 | $4.50 | $45.00 | Official: the International wan2.7-t2v row on Model Studio. Failed requests are not billed.Audio ignored: the row prints no silent / audio split.lotgen's own /pricing still prints Wan 2.7 1080p as unpriced; that is the product catalog, not Alibaba's page. |
| evolink | Kling 3.0 | $0.159/s | 10 | $1.59 | $4.77 | $47.70 | marketplaceMarketplace: the 1080p + audio cell evolink's own Kling 3 page prints, not Kling's list. |
| fal | Kling 3.0 Standard | $0.168/s | 10 | $1.68 | $5.04 | $50.40 | marketplaceMarketplace: fal's own rate for its own key, not Kling's list.Matches Kling's official 1080p audio row on the day checked. |
| Kling | Kling 3.0 | $0.168/s | 10 | $1.68 | $5.04 | $50.40 | Official: 1.2 Units/s × $0.14. Failed generations do not deduct. |
| xAI | grok-imagine-video-1.5 | $0.25/s | 10 | $2.50 | $7.50 | $75.00 | Official: the 1.5 page prints a 480p / 720p / 1080p split under its headline rate; 1.0 has no 1080p cell.Audio ignored: the published rate does not split it. |
| Black Forest Labs | FLUX 3 text-to-video | $0.29/s | 10 | $2.90 | $8.70 | $87.00 | Official: 1 credit = $0.01.Audio ignored: the published table does not split it. |
| Veo 3.1 Standard | $0.40/s | 8 billed (asked 10) | $3.20 | $9.60 | $96.00 | clampedOfficial Gemini rate, audio on by default.8 s billed (asked 10). Veo 3.1 takes 4, 6 or 8 seconds in one take. | |
| BytePlus ARK | Seedance 2.5 | $11.70 / M tokens | 10 | $5.71 | $17.13 | $171.30 | Pixel-area token formula from the BytePlus page, plus the closing frame ARK bills (measured by lotgen).Seedance 2.5, because 2.0 Fast prints no 1080p rate. The 1080p 28% off window ended 2026-09-17 14:00 UTC+8; this is the list.Audio ignored: no published surcharge, and none measured on 2026-08-21.ARK direct is not available to US workspaces; US keys take Seedance on fal or evolink. |
| OpenAI | – | – | – | – | – | – | N/ANo video on this key. GPT Image 2 is stills. |
| ElevenLabs | – | – | – | – | – | – | N/AVoice, not video. |
Cheapest for this spec: Alibaba (Wan 2.7 t2v (International)) at $1.50 a clip.
That is $4.50 per usable shot at 3 attempts, and $45.00 for the whole batch of 30 renders (10 usable shots × 3).
Cheapest means the lowest batch total among rows that (1) have a verified or marketplace-labeled rate and (2) can produce the requested length and resolution without clamping. A clamped row stays in the table and stays out of this callout. Quality differs. This is list price, not a lotgen.ai invoice.
Quality is not in this table. A cheaper second is not a better take. The 3-attempt column is a modeling assumption you set, not a measured keep rate.
How the numbers are built
billed_seconds = clamp(requested_seconds, provider min, provider max)
cost_per_clip = list_rate(resolution, audio, model) × billed_seconds
(or the provider's published unit: tokens, points, credits)
cost_per_usable_shot = cost_per_clip × attempts
batch_total = cost_per_usable_shot × usable_shotsRequested seconds and billed seconds are not always the same. Veo 3.1 on the Gemini / Vertex predict path accepts 4, 6, or 8 seconds (default 8). It does not accept 10 seconds in one take. Omni Flash bills 10 seconds of 720p output whatever duration you ask for (measured by lotgen, 2026-08-21; Google’s public page states about $0.10 per second at 720p and does not print the 10-second floor). MiniMax Hailuo 2.3 at 1080p exists only at 6 seconds. Runway gen4.5 publishes one flat rate and its text-to-video route returns 1280:720, so a 1080p ask is billed at that rate and delivered at 720p. When a row clamps, the table says so, and that row is out of the cheapest callout.
Failed generations: Kling’s official page says failed video generations do not deduct units. Other labs differ. This calculator assumes a successful take. It does not subtract refunds.
What the keys actually are
LOTGEN pastes twelve keys for pictures, clips and voice. That is the list above, in the order the invitation form uses.
Google, BytePlus ARK, fal, evolink, Kling, OpenAI, MiniMax, Black Forest Labs, Alibaba, ElevenLabs, xAI and Runway.
Luma and Sora are models you reach through fal or evolink. Nobody pastes a Luma key or a Sora key here. OpenAI’s key in this product pays for GPT Image stills, not Sora. ElevenLabs is voice, not picture. Those two still appear in the default checkbox list because they are keys; they show as N/A in the table with the reason.
BytePlus ARK is the console that issues the Seedance key, and the name to put on the invoice; the parent group’s name is not the one on the key.
fal and evolink are marketplaces. Their rows are their catalog rates, not the lab’s invoice. Kling 3.0 at 720p on fal currently matches Kling’s official $0.084 / $0.126 per second. That is a coincidence on the day we checked, not a promise.
Get a dated copy, or an invitation
The table above is free. A dated file of the same numbers, and a beta invitation, both go through the invitation form, which asks for your email and which of the four of you you are. This page does not collect an address.
Pick one and the two buttons appear. The form asks the same question, so this only saves you a click.
The invitation form is the gate. A person writes back. If you asked for the dated copy, that reply is where the table goes. There is no second inbox field on this page, and nothing is added to a list you did not submit on the request form.
FAQ
Is this a live quote from lotgen.ai?
No. These are official list prices, dated 2026-09-18, run through the formula above. lotgen.ai shows the exact quote on the button before each take, on your own keys. This page is the public version of that arithmetic.
Why is Veo billed at 8 seconds when I asked for 10?
Veo 3.1 on the Vertex predict path accepts 4, 6, or 8 seconds (default 8). It does not accept 10 seconds in one take. Kling 3.0 can do 10 seconds. The table bills the length the provider will actually run.
Why are Luma and Sora missing from the checkboxes?
They are models, not keys. A fal or evolink key is what you paste; Luma and Sora are rows in those catalogs. OpenAI's key here is GPT Image stills. Sora 2's official API price ($0.10 per second at 720p on 2026-09-18) is on OpenAI's pricing page. It is not a pasteable key on this form.
Are the 3-attempt numbers a measured retry rate?
No. They are a modeling assumption. The default is 3 because that is the concept default. Multiply the one-clip price by however many takes you actually run.
Do fal and official Kling cost the same?
On 2026-09-18, fal's Kling 3.0 Standard text-to-video page listed $0.084 per second with audio off and $0.126 with audio on at 720p, which matches Kling's official 0.6 / 0.9 Units at $0.14 per Unit. evolink's Kling 3.0 page listed a $0.08 per second base. Same model, three bills. The cheapest callout names which key it used.
How do I re-check these prices?
Open the official Kling, Gemini, Vertex, BytePlus ARK, MiniMax, Black Forest Labs, Runway, xAI, fal, and evolink pricing pages linked from the table. Recompute list rate times billed seconds times attempts times shots. If a cell moved, the prices-checked-on stamp moves with it.
Related on this site: how a render is quoted before it runs · Kling vs Veo cost · Kling AI pricing explained · seats and pricing
Official pages read for the table: Kling · Gemini · Vertex · BytePlus ARK · fal · evolink · MiniMax · Black Forest Labs · xAI · xAI 1.5 · Runway · OpenAI · Alibaba · ElevenLabs