VHS to 4K: what a digitized tape can honestly become

Upscaling starts after capture, and capture is not an online step. Nothing here plays a cassette, cleans a video head, sets tracking, corrects a wobbly time base or decides field order. You bring a digital file that is already a faithful copy of the tape's video signal, and everything downstream inherits whatever that file contains. If the only copy you own is a burned DVD from years ago, a phone recording of a television screen, or a compressed upload from a messaging app, that derivative is your ceiling: a model can make it look smoother and clearer, but it cannot restore detail the intermediate copy already discarded.

What a capture should produce is the interlaced standard-definition signal the tape actually carries: a 4:3 picture at 480i in NTSC regions or 576i in PAL regions, saved at a generous bitrate from a player whose heads and tracking are in reasonable condition. What you hand to this tool then has to be a container the browser accepts — MP4, MPEG, MOV, WebM, Ogg, M4V, MKV or AVI — of at most 500 MB per upload, or a public HTTPS video URL, which is the practical route for large captures. A single source is capped at ten minutes of footage, so a two-hour tape becomes a series of segments rather than one submission. Capture once, keep that master untouched, and treat every enhanced file as a derivative you can regenerate.

The order matters more than the settings. Capture, preserve a master copy, correct structural problems such as field order or playback speed, then upscale. Doing it in the other order means paying credits for a file you will want to redo.

VHS is an analog format with a narrow signal budget, and the figures are public. Wikipedia's VHS article states that VHS machines record up to 3 MHz of baseband video bandwidth and 300 kHz of chroma bandwidth, that horizontal resolution is 240 television lines, and that in modern digital terms NTSC VHS is roughly equivalent to 333×480 pixels for luma and about 40×480 for chroma, while PAL VHS is roughly 333×576 and 40×576. The same article lists Super VHS at 420 television lines and notes that slower tape speeds such as LP and EP reduce horizontal resolution further. Vertical detail comes from the television standard rather than the tape: 525 or 625 lines, fewer of them visible after overscan and vertical blanking.

Those numbers are why the phrase VHS to 4K needs care. Capturing a tape at 720×480 or 720×576 pixels is a good idea, because it samples the signal properly, but it does not create detail that the tape never held: the luma figure above is about 0.16 megapixels, and a VHS video's colour detail is far lower still, which is why VHS colour bleeds past edges. Enlarging that to 1920×1080 multiplies pixel count by roughly thirteen, and to 3840×2160 by roughly fifty-two, without multiplying real information by anything like as much. The upscale can still help a modern player, a modern display and a modern encoder produce a smoother picture — it just should not be sold, or bought, as recovered detail.

One practical consequence: sharpening a tape hard enough to look like modern video mostly amplifies tape noise, edge ringing and compression blocks. A restrained target with a mild cleanup usually looks more natural on a television than an aggressive one, and it costs less per minute.

The hosted form sends four things to the provider: the video file or URL, the chosen model, the target resolution tier and the billed duration. There is no field-order option, no aspect-ratio option, no frame-rate option and no audio option in that request. Whatever shape, cadence and sound your uploaded file has is what the model sees, and the upscale happens on that basis.

So interlacing is your problem to solve first. A 480i or 576i capture submitted as-is keeps its combing, and an enhancement pass can sharpen those comb teeth instead of removing them; Wikimedia's interlaced-video article describes combing as the standard artefact of interlaced material in still frames or slow motion. Deinterlace with a method that respects field order, export progressive frames, and only then upload. Decide the frame shape in the same editing step: this pipeline has no aspect-ratio parameter and does not crop, pad or pillarbox for you, so a 4:3 picture stays whatever your file says it is. If the destination is a 16:9 presentation, fit the 4:3 frame deliberately — with pillarbox bars, a blurred background, or a considered crop — before the upscale, not after.

Audio behaves the same way: the request carries no audio settings, so the tape's sound is not part of what this tool manages. If the enhanced video needs its original soundtrack, remux or re-attach the audio after the upscale in your editor. Also treat frame rate as a handover decision — 25 or 29.97 frames per second progressive material should arrive progressive, with any interpolation done deliberately rather than as a side effect.

Both tiers exist and both bill real credits, so choose with the table below rather than by badge. A VHS recording holds roughly 0.16 megapixels of luma detail, 1080p output is about 2.07 megapixels and 4K output about 8.29 megapixels: about thirteen times and fifty-two times the source figure. At the Standard model, 720p and 1080p both bill 9 credits per second, so 1080p is the free upgrade in credits, while 4K bills 33 credits per second — about 3.7 times as much per second for pixels the tape cannot fill.

That makes 1080p the honest default for tape work. Choose 4K when something downstream genuinely requires a 4K frame: a 4K delivery specification, a 4K timeline, a client's technical checklist, or a display pipeline that handles 4K better than 1080p. In those cases you are paying for compatibility and editing headroom, which is a legitimate reason, but it is a different reason from recovering detail. 2K sits between the two at 17 credits per second and is a reasonable compromise for an intermediate master.

There are sources where the calculus shifts. S-VHS recordings, tapes made on good equipment with HiFi audio, and animation with hard black outlines carry more usable structure than an average home recording, so the extra output resolution has more to work with. Even then, test both tiers on one scene and compare before you decide; a tape-level judgement is worth more than a general rule, including this one.

That question is the one most people actually have, and it has a short answer plus a longer one. The short answer: deinterlace the capture, keep it progressive, upload a short scene, upscale it to 1080p with the Standard model, and check the result on a real screen before you process the whole tape. The longer answer is about choosing the scene well and deciding in advance what would make you reject the result.

Pick five to ten seconds from the worst part of the recording, not the cleanest: movement, faces, any on-screen text, and the dropouts, colour noise or timing wobble that made you start this project. Upload it, choose the model and the target tier, and read the credit quote that the preflight shows before submitting — the minimum billed duration is five seconds, so the test is cheap by construction. Write your pass and fail rules down before you look at output. Typical reasons to reject a setting: skin that turns waxy, hair that flickers, straight edges that shimmer, text that changes shape, or grain that turns into moving plastic. If the test passes, encode the rest of the tape into ten-minute segments and process them one at a time, stopping whenever a segment fails.

Keep a short note per tape: capture settings, the scene you tested, the model, the tier and the credits spent. When you come back to a box of twelve cassettes weeks later, that note is what stops you from paying twice for the same discovery. Where a tape is unusually important, run the same scene through two tiers — for instance Standard 1080p against the SeedVR2 model at 1080p, which bills 49 credits per second — and keep the one you can defend on a full-size screen.

Every figure here comes from the published credit table, not from an estimate. At the Standard model, 1080p bills 9 credits per second, which is 540 credits a minute and 32,400 credits an hour of source footage. The same model at 4K bills 33 credits per second, or 1,980 credits a minute, and at 2K it bills 17 credits per second, or 1,020 credits a minute. The SeedVR2 model is a heavier pipeline and bills 49 credits per second at 1080p, which is 2,940 credits a minute. The 720p tier bills the same 9 credits per second as 1080p at the Standard model.

Convert those to the yearly plan you would actually buy. Creator is $238.80 for 28,000 credits, or about $0.0085 per credit, so a minute of 1080p Standard work is roughly $4.60 and an hour is roughly $276. That plan covers about 51 minutes of 1080p Standard footage, about 27 minutes at 2K, about 14 minutes at Standard 4K, or about nine and a half minutes on the SeedVR2 model at 1080p. Starter is $118.80 for 9,600 credits, roughly 17 minutes of 1080p Standard work, and Studio is $598.80 for 80,000 credits, roughly 148 minutes — about two and a half hours of tape.

The useful way to read this table is in real tapes. Two hours of tape at 1080p Standard is 64,800 credits, which is more than two Creator years' worth of credits for one cassette. A ten-tape box at the same settings is over 648,000 credits, well beyond any single plan, which is why the honest advice is to process the tapes you will actually watch rather than the whole shelf, and to test the tier before buying credits in bulk. Model choice moves the price by more than resolution does: SeedVR2 at 1080p costs about 5.4 times Standard at 1080p per second, so reserving it for the tapes that need it changes the budget more than dropping to 720p ever will.

One test scene fits inside almost any purchase, but a family archive does not. A project that runs to several hours of tape at 1080p Standard is a multi-hundred-thousand-credit job, and buying that in small top-ups is both more expensive per credit and harder to plan around. Yearly plans exist for exactly this shape of work: Starter is $118.80 for 9,600 credits, Creator is $238.80 for 28,000 and Studio is $598.80 for 80,000, each billed once with the credits arriving upfront, so you can decide before the project starts how many minutes of tape the budget covers.

The yearly plans also add daily refresh credits — 10 a day on Starter, 25 on Creator and 50 on Studio — which reset every day, are used first and do not come out of the plan's credit balance. For an archive that runs over months rather than in one sitting, a small daily allowance is a naturally paced way to work through a box without paying for a burst you cannot supervise. Do the arithmetic on your own shelf: add up the minutes you genuinely intend to process this year, multiply by 540 credits per minute for 1080p Standard or 1,980 for Standard 4K, and compare the total with 9,600, 28,000 and 80,000.

Two habits keep a plan from being wasted. First, test each new tape's hardest scene at the tier you chose before committing the rest of that tape, because a tape that fails the test should not consume an hour of credits. Second, keep the untouched captures and re-download each approved result while the hosted link is still live; the value of the plan is in finished, stored files, not in submitted jobs.