New · Gemini Omni Flash 1.2: 15s, edit, extend →

Free tool · No sign-up

The Render Ledger

What a second of AI video actually costs: provider rates for 31 video models and 12 image models, read from each provider's own pricing page, plus the retry maths every other comparison leaves out.

Models
43
Priced tiers
113
With audio
11
Verified
2026-10-15

DOI 10.5281/zenodo.24039878

The number nobody publishes

Every AI video cost comparison quotes the same thing: the price of one generation. Nobody ships their first generation. You render, you look at it, you change something, you render again, and every attempt is billed in full whether you keep it or not.

cost of a finished shot = provider rate per second × seconds × attempts until usable

The third term is the one that decides your bill, and it is the one no pricing page mentions. We do not publish a hit rate for each model, because we have not measured one we could defend across the whole roster, and a made-up figure would be worse than none. The calculator below takes that number from you.

Work out your real bill

Set how many attempts it takes you to get a shot you keep. That number is yours, from your own experience. We do not publish one, because we have not measured a per-model figure we could stand over.

Attempts per usable shot

12 shots × 5s × 4 attempts

€25.43

960 credits

Per usable shot
€2.12
Per usable second
€0.424
List price says
€6.36
Attempts you discard
+€19.07

Studio credit cost at the Director pack rate, the lowest per-credit rate we sell.

Video models: cost per second

Provider cost is what the model provider charges, taken from their own pricing page. Credits are what a render costs in AI Video Studio, quoted at the model's default length; what a credit costs is on the pricing page.

ModelTierProvider $/secDefaultCreditsAudioMaxTakes as input
happyhorse-1.1720p$0.145s42No15simage, reference images
1080p$0.185s54
seedance-2.0480p$0.13455s41Yes15simage, reference images
720p$0.30345s91
1080p$0.6825s205
4K$1.565s469
seedance-2.5480p$0.22055s66Yes30simage, reference images, video
720p$0.4735s142
1080p$1.165s320
wan-2.6720p$0.15s25No15simage, video, audio
1080p$0.155s38
wan-2.7720p$0.15s25No15simage, reference images, video, audio
1080p$0.155s38
wan-3.0480p$0.055s13Yes30simage, reference images, video, document
720p$0.15s25
1080p$0.25s50
wan-3.0-prime480p$0.0685s17Yes30simage, reference images, video, document
720p$0.145s35
1080p$0.285s70
minimax-h3768p$0.065s20Yes15simage, reference images, video
2K$0.135s33
4K$0.165s40
minimax-h3-max480p$0.055s16No15simage
768p$0.085s25
1080p$0.165s50
ltx-2.5-pro720p$0.126s54Yes10simage, camera moves
1080p$0.176s77
ltx-2.5-audioStandard$0.13source45Noimage, audio
ltx-2.5-audio-proStandard$0.17source60Noimage, audio
gemini-omni-flash-1.1360p$0.02738s14Yes15simage, reference images, video
720p$0.08198s48
1080p$0.12298s72
4K$0.24578s144
gemini-omni-flash-edit360p$0.0273source9Novideo
720p$0.0819source30
1080p$0.1229source45
4K$0.2457source90
gemini-omni-flash-extend360p$0.02738s14No10svideo
720p$0.08198s48
1080p$0.12298s72
4K$0.24578s144
gemini-omni-flash-upscale720p$0.041source15Novideo
1080p$0.0819source30
4K$0.1638source60
vidu-q4540p$0.0455s14No16simage, reference images
720p$0.0955s30
1080p$0.125s38
2K$0.195s60
4K$0.395s123
flux-3Draft$0.065s18Yes20simage, keyframes
720p$0.175s51
1080p$0.295s87
flux-3-extendDraft$0.125s36Yes20svideo
720p$0.415s123
1080p$0.535s159
flux-3-enhance1080p$0.295s87Notext only
flux-3-max720p$0.175s51Yes20simage, keyframes
1080p$0.295s87
ltx-2.3-reframe720p$0.1source30Novideo
1080p$0.2source60
ltx-2.5-fast720p$0.096s42Yes20simage, camera moves
1080p$0.136s61
1440p$0.196s89
4K$0.36s140
pika-2.2720p$0.045s15No10simage
1080p$0.095s34
pika-2.2-frames720p$0.045s15No25simage
1080p$0.065s23
veed-lipsync-v2Standard$0.07source21Novideo, audio
kling-v4720p$0.1575s55No30simage, reference images
1080p$0.2355s82
kling-v4-edit720p$0.188source65Noreference images, video
1080p$0.282source97
kling-v4-flash720p$0.0945s35No20simage, reference images
kling-v4-flash-edit720p$0.126source45Noreference images, video
omnihuman-v2Standard$0.105source38Noimage, audio

Image models: cost per image

ModelTierProvider $/imageCreditsBatch
flux-2-pro1K$0.0331
2K$0.096
4K$0.2415
flux-2-max1K$0.0751
2K$0.1911
seedream-51K$0.03536
2K$0.0353
4K$0.0353
seedream-5-pro1K$0.067556
2K$0.13510
nano-banana-21K$0.0854
2K$0.128
4K$0.1610
nano-banana-2.11K$0.07944
2K$0.0985
4K$0.17310
nano-banana-pro1K$0.1594
2K$0.159
4K$0.318
gpt-image-2Standard$0.0754
High$0.221514
gpt-image-2.5-flareStandard$0.015734
High$0.05525
Ultra$0.09628
gpt-image-2.5-sunburstStandard$0.015734
High$0.05525
Ultra$0.09628
gpt-image-2.6-flareDraft$0.0277310
Standard$0.04654
High$0.05576
Ultra$0.09679
Max$0.213717
gpt-image-2.6-sunburstDraft$0.0277310
Standard$0.04654
High$0.05576
Ultra$0.09679
Max$0.213717

Questions

What does a second of AI video actually cost?

It depends on the model and quality tier, and it is almost never the number the provider advertises. Across the 31 video models listed here, published provider cost ranges from $0.0273 to $1.56 per second. The figure that matters is that number multiplied by how many attempts it takes you to get a shot you keep.

Why is the real cost higher than the advertised price?

Because you do not ship your first render. Every generation you discard is billed in full. If it takes four attempts to get a usable five-second shot, that shot cost four renders, not one. Most published cost comparisons quote a single generation and stop there, which is why their numbers look lower than your invoice.

Where do these prices come from?

Each figure is read from the provider's own published pricing page and dated in our source data, most recently verified 2026-10-15. Where a provider publishes a token formula rather than a per-second rate, we apply their formula and use the higher of the two when they disagree, so a figure here is never below what you could actually be charged.

Which AI video model is cheapest?

On published provider cost, gemini-omni-flash-1.1 at its 360p tier is the lowest per-render figure in this table ($0.0273 per second). Cheapest per render is not the same as cheapest per usable shot: a model that needs three attempts at a low rate can cost more than one that lands first time at a higher rate.

Which AI video models accept a starting image?

22 of the 31 video models here take a still image as a starting frame, which is how you get a consistent look across shots: generate the frame you want, then animate it. The table's last column lists every input channel each model accepts, including reference images, existing video, audio tracks and explicit camera-move controls.

Can I feed an existing video into these models?

Yes, 15 of them accept video input. Two kinds, billed differently: most take a clip as a REFERENCE alongside your prompt and still bill per second generated, while the dedicated transform tools (reframe, lipsync, audio-to-video) take their length from the source itself, so those show "source" in the Default column, because what you upload decides both the duration and the bill.

Do these prices include audio?

11 of the 31 video models here produce synchronised audio, either through a toggle or natively with no control to switch it off. Audio is included in the per-second rate on those models; there is no separate audio charge.

Render any of these without a subscription

Every model in this table is available in AI Video Studio on pay-as-you-go credits that never expire. Eligible new accounts get free credits topped up daily until their first pack.

Knowing the cost is half of it

This page tells you what a render costs. It cannot tell you which model will get your particular shot right first time. That takes having rendered it. The Shot Ledger is the companion to this table: the same shots rendered across models, with the frame each one produced beside the settings that made it.

Provider costs verified 2026-10-15 against each provider's published pricing. Providers reprice without notice; figures are accurate as at that date. The calculator's euro figures convert credits at the Director pack (€249 / 9,400 credits) rate, the lowest per-credit rate we sell; every other pack costs more per credit, so no figure here understates what you would pay. This page states published prices and documented endpoint capabilities only; it makes no claim about which model produces better results.