What AI Video Actually Costs per Second: Every Major Model, Checked
The provider list price per second for every major AI video model, at 720p, 1080p and each model's top resolution, read from each model's fal.ai pricing page on 26 September 2026. Built from the same table our billing runs on, so it is updated when prices move.
Most AI video pricing is quoted per clip, per credit or per plan, which makes models almost impossible to compare. Underneath all of that sits one number: what the provider charges for each second of video the model generates. This page lists it for every video generation model we run, at the resolutions people actually deliver.
| Model | Maker | 720p class | 1080p | Highest tier | 1 minute at 1080p | Longest take |
|---|---|---|---|---|---|---|
| Veo 3.1 Lite | $0.05 | $0.08 | $0.08 (1080p) | $4.80 | 8s | |
| LTX 2.5 Fast | Lightricks | $0.09 | $0.13 | $0.30 (4K) | $7.80 | 20s |
| Veo 3.1 Fast | $0.15 | $0.15 | $0.35 (4K) | $9.00 | 8s | |
| Wan 2.6 | Alibaba | $0.10 | $0.15 | $0.15 (1080p) | $9.00 | 15s |
| Wan 2.7 | Alibaba | $0.10 | $0.15 | $0.15 (1080p) | $9.00 | 15s |
| Gemini Omni Flash | $0.10 | $0.15 | $0.30 (4K) | $9.00 | 10s | |
| MiniMax H3 Max | MiniMax | $0.08 (768p) | $0.16 | $0.16 (1080p) | $9.60 | 15s |
| LTX 2.5 Pro | Lightricks | $0.12 | $0.17 | $0.17 (1080p) | $10.20 | 10s |
| HappyHorse 1.1 | Alibaba | $0.14 | $0.18 | $0.18 (1080p) | $10.80 | 15s |
| Wan 3.0 | Alibaba | $0.10 | $0.20 | $0.20 (1080p) | $12.00 | 30s |
| Wan 3.0 Prime | Alibaba | $0.14 | $0.28 | $0.28 (1080p) | $16.80 | 30s |
| Flux 3 | Black Forest Labs | $0.17 | $0.29 | $0.29 (1080p) | $17.40 | 20s |
| Veo 3.1 | $0.40 | $0.40 | $0.60 (4K) | $24.00 | 8s | |
| Seedance 2.0 | ByteDance | $0.3034 | $0.682 | $1.5552* (4K) | $40.92 | 15s |
| Seedance 2.5 | ByteDance | $0.473 | $1.164* | $1.164* (1080p) | $69.84* | 30s |
| MiniMax H3 | MiniMax | $0.06 (768p) | n/a | $0.16 (4K) | n/a | 15s |
How these prices were checked
Every figure is the list price on fal.ai, the platform our Studio buys these models through. I opened each model's own pricing page on 26 September 2026 and compared it with the table our billing uses. All but two matched, and both were corrected before this page went live: Seedance 2.0 had crept up a fraction of a cent, and Seedance 2.5 bills 1080p at a higher token rate than 720p, which our table had not reflected.
The table on this page is not typed by hand. It is generated from that same billing table, so when a price changes and we correct it, this page changes with it.
Three things are deliberately left out. Launch discounts are ignored; MiniMax H3 Max is on one until 30 September, and the table shows its standard rate. Transform tools that work on a clip you upload, like reframing or lip-sync, are excluded, because they price your footage rather than new footage. And where a price is an estimate rather than a firm per-second quote, it carries a star.
What a per-second price leaves out
The number is honest, but it is not the whole bill.
You rarely keep the first take. A finished second of footage usually costs several generated seconds once you count the attempts that missed. The cheapest model per second is not the cheapest per usable shot if it needs three times the retakes. Misses cost money too. I covered how to cut that waste in why AI video credits keep getting burned.
Audio can change the rate. On Veo 3.1, turning native audio off lowers the price per second. The table shows the audio-on rate, because that is what most briefs need.
Resolution is a real cost step. Going from 720p to 1080p roughly doubles the rate on several models, and on Seedance it does more than that. Draft at the lower tier. Spend on the take you will actually deliver.
Take length caps the job. The longest-take column matters as much as price. A model that stops at 8 seconds means more cuts, and every cut is another generation.
Reading it for a real job
The price ladder is not a quality ladder. It never was. The expensive rows are not simply better; they are better at particular things, like long single takes or tight prompt adherence. The cheap rows are where you test framing and motion before committing.
So I read the table in two passes, one for each job. The 720p column is the cost of finding the shot. The 1080p column, multiplied by your retakes, is the cost of delivering it. For a full comparison of what each model is good at, see the best AI video generator comparison, and for what a finished production costs end to end, the AI video cost guide.
Where our Studio sits
We buy every model in this table through fal and sell access in credits, so the Studio is not the cheapest way to pay for a second of video. Nobody reselling a model can be. What it gives you is every one of these models from a single balance, with no subscription and credits that never expire, which matters more than the per-second rate once you work across several models. Current credit prices are on the pricing page, and every model here runs in the Studio.
Using this table
You are welcome to reference or quote it. Please credit it as "AIVideos.eu, AI video cost per second" with a link to this page, and include the checked date shown under the table, since these prices move. We re-check them whenever a model's price changes and on a regular review.
Generate cinematic AI video, from €15
31 AI models. No subscription. Buy credits, generate on demand, own the results outright.
Also on Android. The AI Videos Studio app, with the same account and credits.
