Output quality

    Upscaling

    Also called Super-resolution, Enhance.

    Upscaling raises the resolution of media you already have, inventing plausible detail rather than recovering detail that was never recorded.

    Traditional resizing spreads the pixels you have across a bigger canvas, which is why it looks soft. A generative upscaler instead predicts what the missing detail should have been — pores, fabric weave, individual leaves — and draws it in. The result is genuinely sharper and genuinely invented, and both halves of that sentence matter.

    So it is excellent at texture and unreliable at information. Skin, foliage, stone and cloth upscale beautifully. Small text, faces at a distance and logos are where it fabricates: the model produces something letter-shaped where letters were implied, and it will not be the right word.

    On video there is the extra requirement that the invented detail agree across frames, or the clip sparkles. This is why video upscaling is a distinct job with distinct models rather than an image upscaler run over every frame.

    In practice

    • Two passes of 2x usually beat one 4x pass for control.
    • Upscale last, after grading and edits, so you are not enhancing artefacts you are about to remove.
    • Check text and faces at full size — that is where fabrication shows.

    The mistake to avoid

    Upscaling a heavily compressed source. The upscaler treats compression blocks as detail and enlarges them faithfully.

    Where you will run into it

    Related terms

    The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.