AI video upscaler: raise a clip's resolution 1x, 2x or 4x with Topaz

This is resolution scaling, done by a dedicated AI upscaling model rather than simple pixel interpolation — Topaz's video models are built specifically to add plausible detail when enlarging footage, rather than just stretching existing pixels. Pick 2x and a 640×360 clip becomes 1280×720; pick 4x and it becomes 2560×1440. The 1x option is a special case: it keeps the original resolution, which is only useful if a future version of this pipeline adds enhancement-only processing — right now, choosing 1x on VdoBloom simply reprocesses the clip at its existing size.

It does not composite, edit, or otherwise change the content of the video — no object removal, no background change, no re-timing. It also is not a fix for heavily compressed or badly shot source footage: an upscaler can sharpen and enlarge detail that is present, but it cannot recover information that a low-bitrate, motion-blurred or badly exposed source video never captured. Expect a cleaner, larger version of what you fed it, not a fundamentally different clip.

There is no sound-related processing either — this tool only touches the picture. If your source clip has audio, it passes through the upscale, but nothing about the audio quality is affected by this tool.

The request path is simple: VdoBloom uploads your file (or passes through the URL you gave it) to Kie.ai, along with the chosen factor, and Kie.ai runs Topaz's video upscaling model on it. The backend then polls for the result, with an outer 30-minute timeout — if the job has not finished by then, it is reported as failed and any credits already deducted are refunded.

Because pricing follows the source clip's length, VdoBloom reads the video's duration locally in your browser the moment you choose a file (from the file's own metadata) and sends that alongside the upload, so the price you see before generating matches what is actually charged. If a duration cannot be read — for example, when you supply a URL rather than a file — the backend falls back to treating it as a 10-second clip for pricing purposes.

The choice between 1x, 2x and 4x has no effect on which model runs — it is the same Topaz model receiving a different scale-factor instruction each time, which is why the interface for this tool is just three buttons rather than a settings panel.

The clearest case is footage shot or sourced at a lower resolution than you now need it at: an old 480p or 720p product clip that needs to look sharp in a modern ad slot, a screen-recorded demo that needs to fill a 1080p or larger frame, or a short clip pulled from an older camera or phone that needs to sit next to newer, higher-resolution footage without looking soft by comparison.

Because the input budget is small (10MB), the more realistic uses are short-form: a single product shot for a marketplace listing, a short social clip, a few seconds of a scene that needs to be enlarged for a bigger frame — not upscaling an entire finished video end to end.

Every figure below comes from the same pricing function and validation code that actually runs the job — the 30-credit minimum and the two per-second rates are enforced server-side, not just estimated in the UI.

Because pricing is per second, trim a clip down to just the portion you need before upscaling rather than sending the whole take — you pay only for what you upscale, and shorter clips also stay comfortably inside the 10MB upload cap. If your source file is close to that limit, compressing it slightly usually costs less visible quality than being forced to cut the clip shorter.

2x and 1x are billed at the same 3.2 credits/second rate, so there is no cost reason to pick 1x over 2x if you actually want a larger frame — 2x gives you double the width and height for the same per-second price. 4x is billed at 5.6 credits/second, 75% more per second than 1x/2x, so it is worth reaching for only when the destination genuinely needs four times the pixel count rather than as a default.

Since credits are refunded automatically on a failed job, there is no cost risk in testing a short clip first before committing a long one to the same factor.