AI Image Upscaler

Enlarge a small photo or illustration with the compact Real-ESRGAN super-resolution model, locally.

Choose one PNG, JPG or WebP. Check its dimensions before starting. Choose 2× or 4×. Compare the result with the original at full size. Download the PNG copy. Your original file is unchanged.

Drop files here or choose below

Self-made sample files. Processing still runs locally.

PNG / JPG / WebP · One file · 10 MiB per file · 1024 px per side · 1,000,000 pixels · 2× / 4× PNG output. First-use model download, plus engine: 4.72 MiB.

    Help and limits

    What does this tool do?

    Enlarge a small photo or illustration with the compact Real-ESRGAN super-resolution model, locally. It predicts new RGB detail rather than merely stretching existing pixels; it cannot recover facts lost from the original.

    Step by step

    1. Choose one PNG, JPG or WebP. Check its dimensions before starting.
    2. Choose 2× or 4×. Compare the result with the original at full size.
    3. Download the PNG copy. Your original file is unchanged.

    Choosing the settings

    OptionWhat changes
    Options and tradeoffs4× is the model’s native output. 2× downsamples that result and uses the same inference work. The original alpha coverage is resampled separately; it is not invented by the model.
    Formats and limitsPNG / JPG / WebP · One file · 10 MiB per file · 1024 px per side · 1,000,000 pixels · 2× / 4× PNG output. First-use model download, plus engine: 4.72 MiB.

    Example and expected result

    A 256 × 192 image becomes 1024 × 768 at 4×, or 512 × 384 at 2×. These are exact output dimensions, not a promise of better detail.

    If it is not working

    • Choose a smaller image if it exceeds the cap. If engine loading fails, check your connection and retry. Cancel a slow run before trying another file.
    • Inspect faces, fine text and repeated edges. AI can invent textures and change small details. Do not use it as evidence that missing text or features existed.

    Formats and limits

    PNG / JPG / WebP · One file · 10 MiB per file · 1024 px per side · 1,000,000 pixels · 2× / 4× PNG output. First-use model download, plus engine: 4.72 MiB.

    Inspect faces, fine text and repeated edges. AI can invent textures and change small details. Do not use it as evidence that missing text or features existed.

    FAQ

    What does it do?

    Enlarge a small photo or illustration with the compact Real-ESRGAN super-resolution model, locally. It predicts new RGB detail rather than merely stretching existing pixels; it cannot recover facts lost from the original.

    Which option should I choose?

    4× is the model’s native output. 2× downsamples that result and uses the same inference work. The original alpha coverage is resampled separately; it is not invented by the model.

    Why is it not working?

    Choose a smaller image if it exceeds the cap. If engine loading fails, check your connection and retry. Cancel a slow run before trying another file.

    What should I check in the result?

    Inspect faces, fine text and repeated edges. AI can invent textures and change small details. Do not use it as evidence that missing text or features existed.

    Is my file uploaded?

    No. Your files and text are processed locally in this browser, without uploading your input. Engines and models download when needed. Clearing or leaving the page discards the current session; download or copy any results you want to keep.