When you publish a video to Reels, TikTok or YouTube, the platform does not stream the file you uploaded. It decodes it, builds its own copies at the sizes it needs, and encodes each copy inside its own bitrate budget. What your viewer sees is the platform's encode, not yours.
That single fact explains most of the "why does my video look soft after upload" complaints. The file was sharp on your laptop because your player was decoding your file. The moment the platform re-encodes it, the encoder has to throw information away, and whatever it throws away is what your viewer sees.
So the practical question is not "what resolution did I upload". It is "how much real detail did the encoder have to work with, and how much noise did it have to carry".
Your upload is not the file people watch
Platforms do not stream one master. They produce a ladder of versions so the app can pick a size that fits the connection, and the phone in a lift gets the smallest one. Two encodes therefore stand between your edit and the viewer: the platform's encode of your file, and its re-encode for a slower connection. A soft source gets softer at every step.
This is why "upload the highest quality file you have" is better advice than "upload the largest file you have". They are not the same instruction. A large file full of compression blocks and grain gives the encoder more work and no more detail.
What the encoder actually spends bits on
An encoder allocates its budget to whatever changes and is complicated: motion, texture, edges, and noise. Noise is the expensive one, because it is different in every frame and does not compress. A source that is already soft and grainy gives the encoder two problems at once: little real detail to preserve, and a lot of noise to carry.
That is the whole reason people upscale before publishing. A cleaner source, with grain reduced and edges rebuilt, gives the encoder a smaller problem. The platform is not tricked into a higher bitrate. The encoder simply has less junk to encode with the same budget.
When uploading above delivery resolution helps
If the platform shows your clip at 1080p and you upload 1080p, that delivery copy is a first-generation encode of your file. If you upload 2K or 4K, the platform still shows 1080p, but that copy is downscaled from a larger source. Downscaling averages pixels, and averaged pixels are usually cleaner than a single-pass encode of a soft original.
This is a head start, not a guarantee. It helps when the source genuinely has detail to carry. It does not help when the extra pixels are invented, because invented detail is smooth, and smooth detail survives a downscale no better than the soft original did.
When 1080p is enough
Most vertical social video is watched on a phone, in a feed, at arm's length. In that setting a good 1080p upload is usually the right target, and paying for 4K is paying for pixels the screen will discard.
| Your situation | Sensible publish target | Is an upscale worth paying for? | Why |
|---|---|---|---|
| Phone-shot vertical clip, one subject, good light | 1080p | Usually not | The source already has the detail the feed will show |
| Old 480p or 720p clip for a story post | 1080p | Yes, at 1080p | Rebuilding the source helps; 4K adds nothing to a phone screen |
| Reel that is also a paid client deliverable | 1080p publish, 2K master | Yes, at 2K | The master needs headroom for a future edit |
| YouTube upload that will be watched on a TV | 2K or 4K | Yes, at the target you will actually watch | A large screen shows the difference a phone hides |
| 4K source cropped to a vertical | 1080p | No | The crop already discards most of the frame |
| Fast motion, handheld, heavy grain | 1080p | Test first | Motion and grain are where upscalers fail visibly |
The credit cost of a test versus the final
SeedVR2 bills video by the second, with a five-second minimum and a ten-minute maximum per clip. On the standard model the published rates are 9 credits per second at 720p, 9 at 1080p, 17 at 2K and 33 at 4K. The Ultimate model is 131 credits per second at 4K.
| Clip | 720p | 1080p | 2K | 4K |
|---|---|---|---|---|
| 5s (minimum) | 45 | 45 | 85 | 165 |
| 15s | 135 | 135 | 255 | 495 |
| 30s | 270 | 270 | 510 | 990 |
| 60s | 540 | 540 | 1,020 | 1,980 |
At the $29.99 / 1,500 credit pack a credit is about two cents. That makes a 30-second 1080p run about $5.40, the same clip at 2K about $10.20, and at 4K about $19.80. On the Creator subscription at $39 for 3,500 credits a credit is closer to 1.1 cents, so the same three runs are roughly $3.00, $5.66 and $11.00.
The test is cheap. A five-second 720p sample is 45 credits, under a dollar at pack rates, and a new account with a verified email gets one standard 720p run up to five seconds free. That free run is the most useful thing here: it tells you whether the model keeps a face and holds motion before you spend anything on the final.
A 30-second Reel, priced three ways
Take a 30-second vertical clip and run it three ways.
1080p is 270 credits, about $5.40 at the 1,500 pack. 2K is 510 credits, about $10.20. 4K is 990 credits, about $19.80. The 4K run costs 3.7 times the 1080p run for a file that a phone feed will downscale back to roughly the size of the 1080p one.
Estimate the credits for your own job
Pick a resolution and length. The calculator uses the same credit rates as checkout, then shows which pack covers the job.
PricingPage.pricingCalculator.title
PricingPage.pricingCalculator.description
PricingPage.pricingCalculator.estimateTitle
PricingPage.pricingCalculator.creditsValue
PricingPage.pricingCalculator.perVideoSummary
- PricingPage.pricingCalculator.timeLabel
- PricingPage.pricingCalculator.timeValue
- PricingPage.pricingCalculator.sizeLabel
- PricingPage.pricingCalculator.sizeValue
PricingPage.pricingCalculator.videoTimeEstimateHint
PricingPage.pricingCalculator.videoFootageScale
PricingPage.insufficientCredits.failedTaskRefund
PricingPage.pricingCalculator.deliveryHint
PricingPage.pricingCalculator.recommendedPackage
$9.99
PricingPage.pricingCalculator.recommendedPackageHint
The Ultimate model at 131 credits per second is a different purchase. Thirty seconds is 3,930 credits, about $78.60 at pack rates. It exists for work where the clean-up is the product: a paid restoration, or footage with faces that must survive a close look. It is the wrong default for a feed post.
Where the money should go: test, then commit
The pattern I would use on any clip where a face matters:
- Free 720p trial, or a 45-credit 720p run on five seconds.
- If the face and motion hold, a 135-credit 1080p run on fifteen seconds, because motion artifacts do not always show in the first five.
- Only then the final at the target you chose.
That is 180 credits, about $3.60, spent before the final. It is cheap insurance against a long 4K job that has to be thrown away because the model waxed a face at second twelve. Note that 720p and 1080p cost the same 9 credits per second, so once the free trial is gone there is no reason to test below the resolution you intend to deliver.
Why a 4K upload can look worse
More pixels is not automatically a better encode. Three things can go wrong when you push a soft source to 4K before publishing.
The model invents detail. Invented detail is consistent from frame to frame, which is what makes it look plausible on a still and wrong in motion. The encoder then spends bits preserving detail that was never in the scene.
Sharpening halos. A hard edge against a flat background gets a bright rim. At feed size it reads as "crisp". On a larger screen it reads as a mistake.
A 1080p pass that leaves the grain alone often survives the platform encode better than a 4K pass that removed it.
The vertical-crop habit
If you shoot 16:9 and publish vertical, crop first, then upscale, and check what is left. A 4K UHD frame cropped to 9:16 is about 1,215 pixels wide. That is already wider than a 1080p vertical delivery, so the crop does not need an upscale at all.
Order matters for cost too. Upscaling the full 16:9 frame and then cropping pays 4K credits for pixels you delete. If the vertical cut is the deliverable, cut it first, then price the smaller frame.
A delivery checklist
- Keep the source. Never publish an upscale and then delete the original.
- Do not upscale an upscale. Every extra pass adds invented detail that the next encoder treats as real.
- Fix the source before the upscale if you can. A clean export at a higher bitrate beats any repair.
- Compare at playback size. A clip judged at 200 percent on a monitor fails for reasons your viewer never sees.
- Price the run before you upload. The upscale calculator and the video upscaler both quote credits from the real duration.
Limitation
This post cannot tell you what any specific platform will do to your file. Reels, TikTok and YouTube each run their own encoder ladder, their own bitrate caps and their own downscaling, and those settings change without notice. I have not run your clip through any of them, and I do not claim a measured quality gain from uploading above the delivery resolution.
What I can defend is the cost side. The rates are published in the credit cost guide, the 4K cost breakdown, and the live pricing page, and the upscaler quotes the exact number after it reads the file. Whether the platform's copy of your upload looks better is a question only your own upload can answer.