Why Is AI Video Billed by the Second?
Drag the slider: 10 seconds of video = 240 coherent frames + physics that hold + audio in sync. A save-money order comes with it
THE QUESTION THIS PAGE ANSWERS
ANSWER FIRSTWhy Is AI Video Billed by the Second?
Drag the slider: 10 seconds of video = 240 coherent frames + physics that hold + audio in sync. A save-money order comes with it
Read cost as a shape, not a single number. Break a request into input, output, retries, tools, and waiting time. The shape of usage usually tells you which design choice is expensive and where a smaller change can help.
Measure one real request before you optimize an imagined average.
A cheaper call that quietly creates more retries, latency, or review work.
Because a video isn't "one image" — it's twenty-plus images per second that all have to stay coherent, plus synchronized sound and physically plausible motion. The output volume is dozens of times an image's, and the difficulty rises much more than dozens of times.
Frames have to "remember each other"
The character's clothing color, the cup in the background, the direction of the light — no frame is allowed to change its mind. The model has to keep hundreds of frames "in mind" and aligned with each other — far harder than painting hundreds of independent pictures, and the reason characters in early AI videos would "change faces" mid-walk.
Physics has to hold up
Apples must fall, water must flow downhill, hair must move with the wind. The model has to "intuit" the laws of physics from massive amounts of video to avoid giving itself away — this ability burns the most training cost.
Sound has to match the lips
The newest generation of models (Veo 3, Sora 2, and others) generates dialogue and sound effects directly — lip movements, footsteps, and ambient sound all synced to the picture. That's doing the video job and the audio job at once, which is why "with sound" costs noticeably more than "silent."
The complete interaction cost of “Drag It · How Many "Frames" Hide Behind Your Video”
“Because a video isn't "one image" — it's twenty-plus images per second that all have to stay coherent , plus synchronized sound and physically plausible motion.” is a reminder that AI cost is not one price multiplied by one call. Input, output, retries, tools, waiting time, and human cleanup together decide what a task really costs.
Find what the bill repeats
The key variables behind “The character's clothing color, the cup in the background, the direction of the light — no frame is allowed to change its mind .” are usually repeated context, oversized output, retries after failure, and calls that do not produce useful progress. Removing wasted Tokens can reduce cost, latency, and concurrency pressure at the same time.
- Video is priced by the second : about $0.20–0.75 per second in this example — 10 seconds costs thousands of chats
- Why it's expensive : twenty-plus frames per second × frame-to-frame coherence × plausible physics × synced audio
- The money-saving order : polish the script in text → lock the visuals with images → generate video last
A cheaper call can make the whole workflow more expensive
Start with “The newest generation of models (Veo 3, Sora 2, and others) generates dialogue and sound effects directly — lip movements, footsteps, and ambient sound all synced to the picture.” and keep a small table for input, output, retries, tools, and human review. Compare quality before and after optimizing instead of looking at one price in isolation.
From “Drag It · How Many "Frames" Hide Behind Your Video” to “Where the Cost Comes From · It's Not Just "More Images"”
“Drag It · How Many "Frames" Hide Behind Your Video” grounds the problem in “I want to generate a video, length: 5 s — Frames to generate (at 24 fps) — Chats the same money would buy — Rough cost (USD) Pricing is an order-of-magnitude example based on mainstream video model APIs as of A…”. “Where the Cost Comes From · It's Not Just "More Images"” then moves it toward “The character's clothing color, the cup in the background, the direction of the light — no frame is allowed to change its mind . The model has to keep hundreds of frames "in mind" and aligned with each other —…”. Together, they show that the lesson is not just a conclusion to remember, but a claim with conditions.
Carry the judgment into the next situation
When analyzing cost, map the complete interaction first, then find repeated input, wasted output, and retries. A cheap individual call does not make the whole task cheap.
- “Drag It · How Many "Frames" Hide Behind Your Video”: I want to generate a video, length: 5 s — Frames to generate (at 24 fps) — Chats the same money would buy — Rough cost (USD) Pricing is an order-of-magnitude example based on mainstream video model APIs as of A…
- “Where the Cost Comes From · It's Not Just "More Images"”: The character's clothing color, the cup in the background, the direction of the light — no frame is allowed to change its mind . The model has to keep hundreds of frames "in mind" and aligned with each other —…
- “The closing point”: The trend is your friend : prices drop every year — today's "expensive" is temporary
The final “The closing point” brings the discussion to “The trend is your friend : prices drop every year — today's "expensive" is temporary”. The useful thing to carry forward is knowing which judgments must be revisited when input, scale, or risk changes.
✅ What this page wants to share with you
- Video is priced by the second: about $0.20–0.75 per second in this example — 10 seconds costs thousands of chats
- Why it's expensive: twenty-plus frames per second × frame-to-frame coherence × plausible physics × synced audio
- The money-saving order: polish the script in text → lock the visuals with images → generate video last
- The trend is your friend: prices drop every year — today's "expensive" is temporary
I turned one judgment from this article into a small experiment I could run today. Knowing what to observe next is more useful than simply remembering the conclusion.
After reading this, I first looked for the conditions behind the idea instead of copying the method into a project. That order made the later trade-offs much clearer.
When this judgment reaches real work, which constraint should be added first? I am curious which step matters most between reading and the first practical attempt.
No discussion on this article yet.