TL;DR. For 100 clips of 8 seconds (800 s of output), list prices on 20 Nov 2026 range from about $10 to $113 for Grok Imagine Video 1.5 (720p), $19 to $120 for Veo 3.1 Fast (1080p), $83 to $378 for Seedance 2.5 (720p) and $74 to $126 for Kling 4.0 (720p). The same model can cost 2x to 10x more depending on whose API you call. And the per-second price is only half of the bill: what you pay per clip you actually keep depends on your retake rate and on whether the provider bills failed jobs.
Disclosure: we run Turbogen, one of the services in the tables below, and our column comes out lowest on all four models. That is exactly why every competitor number links to the provider's own page: check them. Where we lose is not the per-second price but the terms (credit expiry, API access from the $29.99 pack, fixed clip lengths), covered further down.
Method
- Date. Every competitor number was read from the provider's own page on 20 Nov 2026. Links are in the sources list.
- Job. 100 clips, 8 seconds each, 800 seconds of output.
- Resolution. 1080p for Veo 3.1 Fast, 720p for Grok Imagine Video 1.5, Seedance 2.5 and Kling 4.0. These rows are not comparable to each other, only across providers.
- What is counted. Output seconds with audio, text or image input, no reference video. List prices. Temporary promos are shown separately.
- What is not counted. Retakes, top-up bonuses, free tiers, quality. No quality benchmark here.
Price per second
| Model | Official | fal | Replicate | WaveSpeed | Kie.ai | Turbogen (Boost / Max pack) |
|---|---|---|---|---|---|---|
| Grok Imagine Video 1.5, 720p, audio | $0.14 (xAI) | $0.14 | not listed | $0.14 + $0.01 per request | $0.0225 | $0.0133 / $0.0122 |
| Veo 3.1 Fast, 1080p, audio | $0.12 (Google) | $0.15 | $0.15 | $0.15 | about $0.041 ($0.325 per clip) | $0.0262 / $0.0239 |
| Seedance 2.5, 720p | $0.2311 (BytePlus) | $0.473 | $0.2312 | $0.36 (promo $0.324) | $0.315 | $0.113 / $0.103 |
| Kling 4.0, 720p | not listed | $0.157 | $0.157 | $0.157 (promo $0.133) | $0.1425 | $0.101 / $0.092 |
Notes on the cells that need them:
- xAI prices Grok Imagine Video 1.5 by resolution: $0.08 / $0.14 / $0.25 per second for 480p / 720p / 1080p. fal and WaveSpeed use the same rates. Replicate has the model as an image-to-video preview without a public price.
- Kie.ai lists Grok Imagine at 4.5 credits per second at 720p, $0.005 per credit.
- BytePlus bills tokens: $10.70 per million at 480p and 720p. A 720p second at 24 fps is 21,600 tokens, which gives $0.2311.
- fal also bills Seedance 2.5 by tokens ($0.0214 per 1,000) and states "roughly $0.4730 per second" at 720p on the model page.
- Kie.ai prices Veo per video, not per second: 65 credits at 1080p, $0.005 per credit. The per-second figure assumes an 8 s clip.
- Kling official. We did not find a public per-second API price on a vendor page, so the cell stays "not listed".
- Turbogen sells credits. API access starts at the Boost pack: 3,150 credits for $29.99 ($0.0095 per credit). The largest pack (Max) is 9,200 credits for $79.99 ($0.0087 per credit). Grok Imagine Video 1.5 10 s costs 14 credits, Veo 3.1 Fast 8 s costs 22 credits, Seedance 2.5 10 s at 720p costs 119 (65 for 5 s), Kling 4.0 5 s at 720p costs 53. Per-second figures are those prices divided by clip length.
What 100 clips cost
Per-second rate times 800, rounded to the dollar.
| Model | Official | fal | Replicate | WaveSpeed | Kie.ai | Turbogen Boost | Turbogen Max |
|---|---|---|---|---|---|---|---|
| Grok Imagine 1.5 | $112 | $112 | not listed | $113 | $18 | $11 | $10 |
| Veo 3.1 Fast | $96 | $120 | $120 | $120 | $33 | $21 | $19 |
| Seedance 2.5 | $185 | $378 | $185 | $288 (promo $259) | $252 | $91 | $83 |
| Kling 4.0 | not listed | $126 | $126 | $126 (promo $107) | $114 | $81 | $74 |
Four things stand out.
Grok Imagine Video 1.5 has the widest spread. The same 100 clips cost $112 at xAI's own 720p list price and on fal, $113 on WaveSpeed, $18 on Kie.ai and about $10 on Turbogen. This is the model where we are furthest below everyone else.
Seedance 2.5 is where the provider matters most in absolute dollars. fal is about 2x the official rate. Replicate matches the official BytePlus rate to the cent at $185 per 100 clips. Our price is about half of that ($83 to $91), on clips of 5, 10, 15 or 30 s.
Veo 3.1 Fast is the cheap model, and credit-based resellers undercut Google's list price. That covers Kie.ai and us. fal, Replicate and WaveSpeed all sit at $0.15, above Google's own $0.12.
Kling 4.0 has one list price almost everywhere. fal, Replicate and WaveSpeed all show $0.157. Kie.ai and Turbogen are below it.
One caveat on our own column: Seedance 2.5 on Turbogen is sold as 5, 10, 15 and 30 s clips and Grok Imagine as 6, 10 and 15 s clips, so an 8 s clip does not exist there. The numbers above use the 10 s rate (the 5 s clip costs a little more per second). In practice you would generate 5 s or 10 s clips.
Per second vs per usable clip
Nobody ships every clip they generate. If you discard a share r of your output (wrong motion, broken hands, garbled text), the cost of a clip you keep is:
def cost_per_usable_clip(price_per_s, seconds, discard_rate): return price_per_s * seconds / (1 - discard_rate) # Illustration only, the discard rates are assumptions, not measurements cost_per_usable_clip(0.12, 8, 0.50) # Veo 3.1 Fast at Google list: $1.92 cost_per_usable_clip(0.2312, 8, 0.10) # Seedance 2.5 on Replicate: $2.06 At list price a Seedance 2.5 clip costs about 1.9x a Veo 3.1 Fast clip. If the cheaper model makes you throw away half of the shots of a given type and the dearer one a tenth, the gap closes to about 7%. The discard rates above are made up to show the arithmetic. Yours depend on the shot type, so measure them per model on your own prompts before you pick a default.
Discarded clips are always paid, on every provider. A clip that rendered fine but looks wrong is a successful job. What differs is what happens when the job itself fails or is refused.
Who bills failed jobs
Only what the public docs say, quoted or paraphrased, checked 20 Nov 2026.
| Provider | Failed or refused job |
|---|---|
| BytePlus | "You are only charged for successfully generated videos. No fee is charged if generation fails due to reasons such as content moderation." |
| Replicate | "For all models, if a run fails, we don't charge you." Canceling a run on an official model may still be charged. |
| fal | Server errors (HTTP 500+) are never charged. Client errors such as invalid input (422) "may still be charged". Content policy violations are 422 errors. The FAQ does not say whether they are billed. |
| WaveSpeed | Failed requests, system errors and timeouts get an automatic credit reversal. The refund policy page does not mention moderation refusals. |
| Kie.ai | Homepage: "failed tasks are never charged". |
| Not stated on the pricing page we read. | |
| Turbogen | Failed generations are refunded to the balance automatically. |
If your prompts sit close to moderation limits (real people, brands, violence in historical scenes), the moderation row is worth more than a few cents per second. Ask support before you commit volume.
Footnotes that change the bill
- 1080p Seedance 2.5 is roughly 2.5x the 720p price: $1.164/s on fal, $0.90 on WaveSpeed, $0.79 on Kie.ai, about $0.57 on BytePlus ($11.70 per million tokens). Turbogen is the exception: native 1080p is about $0.14 to $0.15/s. Replicate lists 480p and 720p only.
- Kling 4.0 at 1080p: $0.235/s on fal, Replicate and WaveSpeed, $0.2125 on Kie.ai.
- Audio on Veo. Replicate and WaveSpeed show $0.15 with audio and $0.10 without. Google Cloud lists $0.12 with audio and $0.10 without at 1080p. Check which one a headline price refers to.
- Reference video input changes billing almost everywhere: a lower rate, but input seconds are billed too.
- Credit expiry. Kie.ai and WaveSpeed say credits never expire. Turbogen credits expire after 60 days on the API packs (30 days on the entry web-only pack), which hurts if your volume is lumpy. API access starts at the Boost pack ($29.99), so there is no $10 way to test the API.
- Clip length. Veo 3.1 Fast stops at 8 s. Seedance 2.5 and Kling 4.0 go to 30 s in one take.
How long you wait
Price tables hide queue time. We can only publish our own numbers: Turbogen production, 30 Oct to 20 Nov 2026, all users, wall-clock from submit to finished file.
| Model | Median | p90 | Jobs |
|---|---|---|---|
| Veo 3.1 Fast | 218 s | 678 s | 15 |
| Seedance 2.5 | 302 s | 687 s | 46 |
| Nano Banana Pro (image) | 123 s | n/a | 118 |
These are small samples, so treat them as an order of magnitude: minutes, not seconds, with a long tail. We have no comparable data for the other providers, and we would not trust a number we did not measure. If you have yours, see the last section.
Which one to pick
- Seedance-heavy workload: Turbogen if price per clip matters most. Replicate or BytePlus direct if you want the official per-second billing with no packs and no credit expiry, plus a clear "failed runs are free" rule.
- Grok Imagine on a budget: Kie.ai or Turbogen, both a fraction of the official and fal prices.
- Veo-heavy workload on a budget: Kie.ai or Turbogen. Roughly a third or less of Google's list price.
- Contracts, data-handling terms, newest Veo features first: Google directly.
- SDK, queue tooling, many endpoints in one place: fal. Rarely the cheapest on these models.
- Kling 4.0: list price is the same on fal, Replicate and WaveSpeed, so pick by tooling, or go lower with Kie.ai or Turbogen.
- Mixed models plus a web studio on the same balance: that is our case for Turbogen, with the expiry caveat above.
Limitations
- A one-day snapshot. These prices move often, and promos come and go.
- List prices for output only. No retakes, no volume discounts, no top-up bonuses (Kie.ai advertises 10% on larger top-ups).
- No quality comparison. A provider can serve the same model with different defaults.
- The failed-job table reflects public docs, not tested behavior.
- We sell one of the options. Check the linked pages yourself.
Sources (all checked 20 Nov 2026)
- Google: Google Cloud generative AI pricing, Veo 3.1 Fast, video + audio
- BytePlus: ModelArk pricing, Seedance 2.5 token price and billing rules
- fal: pricing (Veo 3.1 Fast), Seedance 2.5, Kling 4.0, FAQ
- Replicate: google/veo-3.1-fast, bytedance/seedance-2.5, kwaivgi/kling-v4.0, billing docs
- WaveSpeed: Veo 3.1 Fast, Seedance 2.5, Kling 4.0, refund policy
- xAI: Grok Imagine Video 1.5; fal Grok Imagine 1.5; WaveSpeed Grok Imagine 1.5; Replicate xai/grok-imagine-video-1.5
- Kie.ai: Grok Imagine, Veo 3.1, Seedance 2.5, Kling 4.0, homepage for the failed-task rule
- Turbogen: our own catalog, turbogen.app/pricing
Add your numbers
Post your provider, model, resolution and real cost per kept clip in the comments, and we will add it to the table with credit. Corrections are just as welcome: if a number here is wrong or already outdated, say which one and link the page.
Researched and drafted with AI agents. Every price was re-read on the linked provider pages on 20 Nov 2026, and the Turbogen team reviewed the result.
