4K is a different per-second rate, not a surcharge
Resolution-banded models publish separate credits-per-second figures per band, so the premium compounds with duration and is not a constant across the catalogue.
Resolution-banded models publish separate credits-per-second figures per band, so the premium compounds with duration and is not a constant across the catalogue.
There is no line item called “4K surcharge” that you pay once. What does exporting in 4K instead of 1080p cost? is a lookup: credits = rate_per_second(resolution) × seconds. You land on one band in the model’s price table. That rate then multiplies every second of the clip. A twenty-second hero at the top band is not a ten-second clip plus a flat 4K fee. It is twenty times the 4K rate.
A lookup, not a multiplier you reuse
Each resolution-banded video model lists its own credits-per-second figures. Some group SD and HD into one price. Others separate them. Several treat HD and 4K as a single band — on those rows, exporting 4K genuinely costs nothing extra. The band labels are the catalogue’s own. Read the row you are actually dispatching.
The worked example on the cost page uses LTX 2.3 Text to Video Fast. Billing type is resolution_based_per_second. Options:
- SD/HD — 4 credits per second
- 4K — 16 credits per second
A ten-second clip is 40 credits at SD/HD and 160 credits at 4K. Same model. Same formula. Different cell. The 4× on this row is this row’s gap, not “the 4K premium” for the whole catalogue. Another model can charge a gentle step between its cheapest and dearest band. A third can make the top band dominate the bill. If you export high-resolution habitually, the model you standardise on is worth more to the balance than any other single decision on that page.
Do not budget “four times 1080p.” Check the matrix. The app quotes the whole-credit total before you confirm.
Duration multiplies the gap
Because the premium is the difference between two per-second rates, it grows with length. On the LTX Fast row the gap is 12 credits every extra second (16 − 4). Six seconds: 24 versus 96. Ten: 40 versus 160. Fourteen: 56 versus 224. Twenty: 80 versus 320. The difference column is itself linear in seconds.
A one-second test at 4K is cheap enough to ignore. A twenty-second hero cut at 4K on a steep model is one of the largest single charges available. Trim first, then raise the resolution. Raising the band on a cut you have not locked is how a scratch take becomes a 4K invoice.
This is the same shape as other time-priced meters: the unit is a second, at the selected resolution. Native audio, where a model prices it, is a different switch on the same kind of rate — not this page. Storage and delivery are not in the credit. Credits cover the render.
Capability rank is not this formula
/best/best-4k-ai-video-model is a different query. It filters video generators whose maximum output resolution is 4K and ranks them by leaderboard position, then job cost. It answers “which 4K rows exist and who is ahead on the board.” It does not answer “what does this extra band cost on the model I already picked.”
The cheap sibling is a price ranking of that same 4K pool. Two pages over one set, different sorts. Neither one replaces the per-second lookup. A model can list 4K as its ceiling and still bill a modest band jump, or a brutal one. Read both: capability on /best, formula on /cost.
Upscale a video to 4K is a third transaction: upscale_video on a file you already have, priced by the upscale model, not by this generate formula. Sometimes 1080p plus one upscale beats one top-band generate. Sometimes it does not. Compare the two totals against your model’s multiplier. Upscaling sharpens what is there; it does not recover detail the generate never made. Artifacts get sharper too.
When the top band buys nothing
Vertical social feeds re-encode the upload. Paying the 4K cell for a clip that will be downscaled on TikTok or Reels buys a master you will not ship. Generate at the band you will deliver. Keep 4K for plates that will actually be shown at 4K — a site hero, a deck, a screen that holds the pixels.
On many models 1080p is not more expensive than 720p, because SD and HD share a band. The extra spend is the jump to the dearest label, often 4K. If HD and 4K share a band, the “premium” is zero on that row. That is not a promotion. It is how the matrix published.
Plan fit is the same arithmetic as the cost page: a month of credits buys fewer ten-second clips at the top band than at SD/HD, on the same model, because the rate switched. Count jobs at the band you will actually export.
Do not:
- Treat 4K as a checkbox surcharge you pay once per project.
- Standardise on the steepest multiplier because the board liked that model at 1080p.
- Upscale and generate at 4K for the same take “to be safe.” One of those jobs is enough; both bill.
The useful finding is that there is no such thing as “the 4K premium.” There is a cell in a table, then multiplication by seconds.
FAQ
Is 4K always four times the price of 1080p?
No. LTX 2.3 Fast happens to be 4 credits/sec at SD/HD and 16 at 4K — a 4× on that row. Other resolution-banded models publish different gaps, including rows where HD and 4K share a band. Read the model’s matrix.
Does a discarded 4K take still cost the 4K rate?
Yes. The charge attaches to the render, at the band you asked for. Test at the cheap band, then raise resolution on the keeper.
Is it cheaper to render at 1080p and upscale?
Sometimes. That is one standard-band generate plus one upscale charge, against one top-band generate. Compare both quotes. Upscale is not this formula.
Why does the cost page talk about a lookup instead of a surcharge?
Because you are not paying a flat 4K fee on top of 1080p. You select a different per-second rate, and that rate multiplies duration. The premium compounds with length rather than being paid once.