It raises the resolution of an existing video while rebuilding the detail that was never captured or was lost to compression. Unlike scaling in an editor, which simply spreads the same pixels over a larger frame, the model predicts plausible fine structure — hair, fabric weave, text edges, skin texture — from what it learned about how real footage looks before and after degradation.

MP4 and MOV, up to 50 MB per file. We read the real duration and resolution of the file on our servers before the job starts, so a clip that cannot be read is rejected before any credits are spent.

By the length of the source video, counted in whole seconds — a partial second at the end is not charged, and every job bills at least one second — at the per-second rate shown next to the cost in the form. Upload your clip first and the panel shows the exact figure before you submit. Failed jobs are refunded automatically.

Choose the size your delivery needs. 2K uses the 1440p tier. Sizes below the source are hidden; a matching size remains available for enhancement. Exact dimensions follow the source aspect ratio.

No. The upscaler is source-preserving: it takes no prompt and cannot add, remove or re-time anything. Duration, framing, cuts and motion come back unchanged; only resolution and clarity differ.

Yes, and it is one of the most useful cases. Many models cap output at 720p or 1080p. Generate at the model's native resolution, pick the result from My Creations, and upscale it afterwards — usually cheaper than paying a premium tier for a higher-resolution generation.

Time depends on source duration, output size and the provider queue. The job continues after you leave; collect the finished video from your history.

No. The upscaled MP4 comes back clean and ready to publish, at the same duration and aspect ratio as the file you uploaded.

No. Everything runs on our servers, so the work does not depend on your machine's GPU. Upload from a browser on Windows, Mac or a phone, close the tab if you like, and collect the result from your history later.

Footage that is soft or small but otherwise clean. Heavy compression blocking, severe motion blur and extreme low light limit what any model can recover, because the detail is not merely hidden — it was never recorded. Start from the highest-quality master you have rather than a re-exported or re-uploaded copy.