Upscale 480p to 1080p without overclaiming

480p is a family of progressive-scan resolutions rather than one size. Wikipedia's 480p article describes it as carrying a vertical resolution of 480 pixels, ordinarily with a horizontal resolution of 640 pixels at 4:3 or 854 pixels for approximately 16:9, and notes that standard definition has always been a 4:3 picture at 720 × 480 and 60 Hz in NTSC regions and 720 or 768 × 576 in PAL regions, where a 15.7 kHz horizontal scan rate means those modes are interlaced rather than progressive.

In practice 480p arrives from a few recognisable places: early digital cameras and camcorders, older phone footage, exports made for a small screen, downloads that were transcoded down, and screen recordings captured at a small window size. Each has a different failure mode. A native 480p capture is soft but clean; a downloaded transcode carries blocking and mosquito noise from the encode; a screen recording of a player carries both, plus whatever scaling the player applied.

That distinction decides what the upscale can do. A clean 480p capture is the good case: the model can reconstruct edges and texture plausibly because the source is internally consistent. A re-encoded 480p file is harder, because the model is being asked to work from a picture that already contains an encoder's guesses, and enlarging it enlarges those guesses too.

A 640 × 480 frame contains 307,200 pixels, or about 0.31 megapixels. A 720 × 480 frame contains 345,600, or about 0.35. A 1920 × 1080 frame contains 2,073,600 pixels, about 2.07 megapixels, which is roughly six times the smaller 480p frame and about six times the wider one. A 3840 × 2160 frame contains 8,294,400 pixels, about 8.29 megapixels, roughly twenty-four times a 720 × 480 source.

Those multipliers are the point. Upscaling to 1080p means the model has to produce six output pixels for every input pixel, so most of what you see in the result is reconstructed rather than recorded. That is not automatically bad: a modern display, a modern encoder and a 1080p delivery timeline all handle a 1080p frame better than a 480p one, so there is a real presentation benefit. It is simply a different benefit from recovering detail.

Compression complicates the arithmetic in one direction only. If the 480p file is a low-bitrate encode, its pixels are already an approximation, and enlarging it enlarges blocking and mosquito noise before any reconstruction reduces them. Wikipedia's compression artifact article notes that block artifacts are most apparent in flat areas and that video artifacts persist across decompressed frames and travel with the optical flow, which is why a chroma-subsampled, heavily compressed source is the least rewarding starting point.

Choose five to ten seconds that contain the hardest material: a dark scene with visible noise, a fast pan across texture, on-screen text, and any face. Upload it, pick the model and the output tier, and read the credit quote before submitting. At a five-second minimum the test is cheap, and it answers the question that matters most — whether 1080p or 4K is the right target for this particular file.

Fix your pass and fail rules before viewing. Reject a setting when text becomes unreadable or reshaped, when a face changes identity, when flat areas crawl, when edges grow haloes, or when the result looks sharpened rather than resolved. Accept it when the frame plays smoothly on your target display, when edges are stable across a pan, and when the reconstructed texture is plausible at normal viewing distance.

Then process the rest in 10-minute segments, because a single submission is capped there and a long clip becomes several jobs. Keep the untouched 480p source, and keep a short note per segment: source, scene, model, tier, credits and verdict. If a segment fails, only that segment needs a re-run at a heavier model, which is where the heaviest models earn their price.

Everything here comes from the published credit table. At the Standard model, 720p and 1080p both bill 9 credits a second, which is 540 credits a minute and 32,400 an hour of source footage; 2K bills 17 a second and 4K bills 33. FlashVSR bills 30 credits a second at 1080p, and SeedVR2, Pro and Ultimate bill 49 credits a second at 1080p.

Against the yearly plans, Creator is $238.80 for 28,000 credits, which is about 10 hours of 1080p Standard work; Starter is $118.80 for 9,600 credits and Studio is $598.80 for 80,000 credits. If you only want to check whether a particular clip is worth processing, the smallest public general pack is $29.99 for 1,500 credits.

Two decisions move the bill more than anything else. Output resolution: because 720p and 1080p cost the same at Standard, staying at 720p saves nothing, while stepping up to 4K multiplies the rate by about 3.7 for pixels the source cannot fill. Model choice: moving from Standard to the heaviest models at 1080p multiplies the rate by more than five, so reserve them for the segments that actually fail at Standard.

Compare the same frame at 100 percent with the untouched 480p source, side by side, on the display you will deliver for. Then step back to normal viewing distance and watch it at speed. Both views matter: the 100 percent crop exposes invented texture and reshaped letters, while the normal-distance view tells you whether the reconstruction holds together as motion rather than as a still.

Pay particular attention to the three things 480p material usually breaks. Fine repeating texture such as fabric or foliage can turn into a pattern that the model invented rather than resolved; small text can change shape between frames; and gradients in skies or dark interiors can develop banding that either softens or crawls, depending on the model. If any of those three changes between adjacent frames, the setting is wrong for that scene.

Keep the comparison short and written: source, timestamp, model, tier, credits, verdict, and the single reason. Because the same clip will look different at 720p, 1080p and 4K for structural reasons rather than quality reasons, notes are the only way to avoid re-running the same test from memory.

A hosted job here sends the file or a public URL, the model, the target resolution tier and the billed duration, and nothing else. There is no aspect-ratio setting, no crop or pad option, no frame-rate option, no bitrate target and no audio option. Whatever shape, cadence and sound your 480p file has is what the model sees, and the output inherits it.

So framing is your decision, made before upload. A 4:3 480p file stays 4:3 unless you letterbox or crop it yourself; a 16:9 presentation built from a 4:3 source needs a deliberate choice — pillarbox, blurred background or a considered crop — taken in the editor rather than as a side effect. Interlaced 480i captures should be deinterlaced to progressive frames first, because an enhancement pass on interlaced material can sharpen comb teeth instead of removing them.

Audio behaves the same way. The request carries no audio settings, so the original soundtrack is preserved and re-attached by you after the upscale. That keeps the pipeline honest: it does exactly one job — enlarge and reconstruct the picture — and it does not quietly rescale, re-time or re-encode anything else about the file.