The honest answer to "how much does 4K upscaling cost?" is: it depends on the length of the clip, the model, the resolution you start from, and whether you pay per job or own the hardware. The useful answer is a table you can apply to your own file. This guide uses the published SeedVR2 credit rates and shows the math for common clip lengths.
The first thing to fix is the unit. A 4K upscale is not priced per file. It is priced per second of source footage, with a minimum charge. That is why a 5-second clip and a 60-second clip can differ by more than ten times even though both become one 4K file.
The published rate card
On the standard SeedVR2 video model, the published rates are:
| Target | Credits per second | 5-second minimum |
|---|---|---|
| 720p | 9 | 45 |
| 1080p | 9 | 45 |
| 2K | 17 | 85 |
| 4K | 33 | 165 |
The 4K rate is 33 credits per second. A clip shorter than five seconds still bills as five seconds, so the smallest 4K job is 165 credits. A 10-second clip is 330 credits. A 30-second clip is 990. A 60-second clip is 1,980.
Those are credits, not dollars. The dollar conversion depends on the pack you buy. At the published $29.99 / 1,500 credit pack, one credit is about two cents. That makes the examples:
| Clip length | 4K credits | Approx. cost at $29.99 / 1,500 |
|---|---|---|
| 5 seconds (minimum) | 165 | about $3.30 |
| 10 seconds | 330 | about $6.60 |
| 30 seconds | 990 | about $19.80 |
| 60 seconds | 1,980 | about $39.60 |
Treat the dollar column as an estimate. Pack prices and promotions change. The live pricing page and the preflight quote are the source of truth.
Why the 4K number is not the whole cost
The credit rate is the visible cost. There are three less visible ones.
Failed jobs. A failed generation should return its credits, but you still lose the time and the review cycle. A provider outage, a content-policy rejection, or an infrastructure error can stop a job after you have already uploaded and waited. The refund policy matters as much as the rate card, because a cheap job that fails twice is not cheap.
The wrong resolution. If the source is a compressed 480p phone clip with no real detail, a 4K target may produce a larger file without adding usable information. You pay the 4K rate for pixels the model had to invent. A 1080p or 2K run is often the better first test. If the result holds up at 2K, then decide whether 4K adds anything.
Your review time. A 60-second 4K job is not finished when the file arrives. You still need to watch it, compare faces, check edges, and decide whether the output is deliverable. That time is real, especially for client work.
2K versus 4K: when the cheaper job wins
The 2K rate is 17 credits per second, roughly half the 4K rate. For a 30-second clip, that is 510 credits instead of 990. The question is whether the difference is visible in your delivery.
Use 2K first when:
- The source is soft, compressed, or below 720p.
- The final delivery is a phone screen, a social feed, or a web player.
- You are testing whether the model preserves identity before spending on 4K.
- The clip has fast motion and compression artifacts that 4K will enlarge rather than fix.
Use 4K when the source has real detail, the delivery asks for a 4K master, and the 2K test already looks clean. The worst outcome is paying the 4K rate for a clip that needed cleaning, not more pixels.
What local GPU upscaling really costs
A local run has no credit meter, but it is not free. You pay for the GPU, the machine, the electricity, and the time. The electricity is usually the smallest part. A 300W GPU running for ten minutes uses about 0.05 kWh; at $0.15 per kWh that is less than a cent. The hardware and the setup time dominate.
That is why the comparison is not "hosted credits versus zero". If you already own a capable GPU and run the same workflow every day, the marginal cost of another local job is small. If you are buying a GPU for one family video, the hosted job is usually cheaper. The break-even point depends on how many jobs you run and how much you value the setup time.
How to estimate your own clip
- Check the source duration in seconds. Round up to the next whole second.
- Apply the five-second minimum.
- Multiply by the per-second rate for your target. Use 33 for 4K and 17 for 2K on the standard model.
- Compare the credit total with the pack price. Do not compare with the largest pack if you only need one job.
- Run a 2K test first if the source is soft or the face matters.
The upscale cost calculator does the arithmetic for you, and the live video upscaler quotes the exact credits after it reads the real file duration.
The hidden cost of a bad output
A cheap upscale that changes a face is more expensive than a slightly higher-priced job that preserves it. For family footage, the face is the product. For product footage, the logo and texture are the product. For client work, a rejected deliverable costs another job, another review, and sometimes a refund.
This is why the first run should be a test, not a batch. Pick one clip, run it at the resolution you can afford, and inspect the details that matter. If the result is clean, scale the workflow. If it is not, change the source, the target, or the model before you spend on a longer clip.
FAQ
How much is a 10-second 4K upscale? 330 credits on the standard model. At the published $29.99 / 1,500 credit pack, that is about $6.60.
How much is a 30-second 4K upscale? 990 credits, about $19.80 at the same pack price.
Is 4K always better than 2K? No. If the source has no real detail, 4K can enlarge artifacts. Test 2K first when the source is soft or compressed.
Do failed jobs cost credits? The system attempts to return credits for provider failures, policy rejections, and infrastructure errors. Check the current refund policy for the exact window and limits.
Can I upscale a clip longer than one minute? The video tool supports longer clips up to the published maximum, and the cost scales per second. A 60-second 4K job is 1,980 credits on the standard model.
Price your own clip: open the video upscaler, upload the file, and read the credit quote before you submit. Start at 2K if the source is soft; move to 4K only when the test output earns it.