No watermark on the free export.Transparent WebM, GIF or green screen, straight out — free exports go up to 1920 px.
By ShinobiTools Team · Last updated: July 2026
Make an old, small or blurry clip four times bigger — 640×360 comes back as 2560×1440. Our upscaler runs frame by frame on our own GPU, your original audio is copied over untouched, and you get one plain MP4 back. No account, no watermark, 20 seconds free.
In the 40 days before this was written, 113 clips went through this upscaler. The average was 12 seconds, and the longest was 2 minutes 36 seconds — well past the free 20-second ceiling, so that one needed a pass. Most people stay comfortably inside the limit: just over half used less than half of the allowance, and the two biggest groups ran 4 to 8 seconds (49 clips) and 8 to 16 seconds (45 clips) — short, specific rescues, not long home movies.
In the same window, 14 of those upload attempts ran longer than the limit that applied to them, and 11 of those came back with the exact message "413 Clip is longer than 20 seconds — please trim" — the free-tier ceiling most people hit. A further 7 landed within 10% of their limit, close enough that a few extra frames would have tipped them over. If your clip is borderline, trimming a second or two off the end first is the fastest way through — the free tier is genuinely built for short rescues, not for upscaling a whole video.
Removing a background is fast. Upscaling is not, and we would rather say so up front than have you stare at a bar. Every single frame goes through the model, so at the free 640×360 input one second of video costs us roughly seven seconds of GPU time. A 20-second clip is about two and a half minutes of waiting. You will see the frame counter tick by (“frame 120/500”) so you can tell it is working, and the queue tells you how many people are ahead of you.
Measured on our own card on 5 August 2026, including decoding and re-encoding: an input of 480×270 takes 148 ms per frame, 640×360 takes 280 ms, and 854×480 takes 545 ms. That is why the input resolution is capped before anything is upscaled — it is the one number that decides everything.
Cutting a background out of a clip is cheap for whoever runs the server. Upscaling is not: it is the one operation where the bill scales directly with the length of your video, because there is no shortcut around running every frame through a neural network. That is why compute time, rather than file size, is usually the first thing a free tier has to limit.
We already own the cards — they are in a machine in a room here, not rented by the minute — so we can be more generous with them. Free gets you 20 seconds, which covers the clips people actually want to rescue: a video from an old phone, a VHS transfer, something downloaded years ago at a size that looked fine on the screens of the day.
It is not unlimited and we will not pretend otherwise. If you need minutes rather than seconds, a pass raises the limit — and that is a real cost you are covering, not an artificial gate.
Free: clips up to 20 seconds. Anything larger than 640×360 is scaled down to that size first and then upscaled 4×, so you get 2560×1440 back. No watermark, no account, full quality.
Leave your email for the free monthly pass and the limit moves to 60 seconds at the same resolution. With a ShinobiTools pass it goes to 3 minutes and the input cap rises to 854×480, so the output is up to 3416×1920 — a widescreen 16:9 clip lands on 3408×1920, because the input is rounded to an even number of pixels first.
Those numbers are not marketing copy: the tool reads them from the server, so what the page promises is exactly what the upload will accept. And the same honesty applies to the pass: a full 3-minute clip at 854×480 is about 40 minutes of processing. That is what you are actually buying — a long slot on a real graphics card, not a faster one.
This is the same upscaling model our sister site PicReviver uses on photos. It rebuilds plausible detail: edges get crisp, compression blocks and mild noise get cleaned up, and text and faces stop looking mushy. It is genuinely good on footage that was small to begin with.
It is not magic and it does not invent what was never filmed. A clip that is out of focus stays out of focus, heavy motion blur stays blurry, and a heavily compressed source will not come back as a studio master. It also works per frame with no memory between frames, so very noisy footage can shimmer slightly.
Upscaling and background removal are separate jobs here. If you want both, run the clip through one and then the other.
Nothing to install and no account. The upload is deleted right after processing and the result is wiped within 45 minutes.
Four times in each direction. 640x360 comes back as 2560x1440, so sixteen times the pixels. With a pass the input cap is 854x480, which gives up to 3416x1920 (3408x1920 for a normal 16:9 clip, since the input is rounded to an even number of pixels).
About seven seconds of processing per second of video, measured on our own GPU. A 20-second clip takes roughly two and a half minutes, plus any queue. The page shows a live frame counter so you can see it working.
About 40 minutes on our GPU, at 545 ms per frame for a 854x480 input. A pass buys you a longer slot on the card, not a faster one, and we would rather you knew that before you started.
Because upscaling is the one job where the cost grows directly with the length of your clip — every frame goes through the model. Twenty seconds is what we can give away per visitor without the queue becoming useless for everyone else.
No. Not on the free version and not on any other version. You get a plain MP4.
Yes. The original audio track is copied straight into the new file without being re-encoded, so it is bit-for-bit what you uploaded.
It sharpens detail and cleans up compression artefacts, and small or soft footage improves a lot. It cannot recover what the camera never captured: out-of-focus and motion-blurred footage stays that way.
Your upload is deleted right after processing and the result is wiped within 45 minutes. See privacy.