- Is video upscale free?
- No — AI video upscale is a Pro feature: every frame runs through a neural network on our own GPU, so a 30-second clip occupies it for 10 to 15 minutes. Image upscaling stays free for everyone at Upscale Image.
- What's the difference between the engines?
- Standard gives the best speed-to-quality ratio on real footage and includes a light denoise. Quality is a heavier network that can recover more detail on sharp real footage at about 8× the processing time, so it is limited to clips of 15 seconds; on soft or heavily compressed sources it looks much like Standard, so try Standard first. Anime is trained on animation and keeps lines clean and colours flat without the halos a photo model adds.
- Should I pick 'Original size, sharpened' or 'Upscaled'?
- Original size is the right choice when the video looks soft or was heavily compressed and you want it crisper without a larger file: the AI rebuilds detail at 2× or 4× and the result is scaled back to the source dimensions. Upscaled is for when you actually need more pixels, for example a 480p clip you want at 1080p.
- How big can the output be?
- Up to 2160 px on the long edge — 4K height for vertical video, about 2K width for landscape. A 1080p clip at 2× therefore comes out at 2160×1214, not 3840×2160. Larger outputs wait for our next GPU.
- How long does video upscaling take?
- Roughly a second per frame at 1080p output and a fifth of that for small clips, so a 30-second 1080p clip takes 10 to 15 minutes and a 4-second clip about a minute. Quality mode is about 8× slower. Slow-motion sources at 120 or 240 fps are delivered at 60 fps, which cuts the work by 2 to 4× and plays back the same.
- Will upscaling add real detail?
- The network rebuilds plausible detail that the source lost — edges become sharper, textures more defined, compression noise is smoothed. It is not magic: heavily blurred or tiny footage stays limited, but the result is dramatically better than a bicubic resize.
- How does this compare to desktop upscalers?
- Desktop tools with a large GPU budget still edge ahead on archival footage and offer more controls. For short clips and most social media content this gives comparable results, runs on our hardware, and needs nothing installed.
