How to turn an image into pixel art without making it blurry

Start with a tightly cropped subject, a 64-pixel output size and 16 colours in the pixel-art converter. Leave dithering off for the first pass. Check whether the subject reads clearly before making it smaller or adding texture.

The tool uses nearest-neighbour sampling for its final reduction and offers whole-number PNG export scales. You do not need to find a separate “nearest neighbour” switch.

Make a useful first version

1. Choose your image. A simple silhouette or close crop gives important features more room than a busy scene.
2. Select 64 × 64 under Output size. Keep aspect ratio fits the whole image inside that box; Crop to fit makes a centred square crop.
3. Choose From this image, set Colours to 16 and Dithering to None.
4. Compare the original and result. Try 128 if eyes, lettering or thin shapes have disappeared.
5. Adjust brightness, contrast or saturation only where it helps. Raising all three is not a required step.
6. Save PNG 1× for further pixel editing, or a larger multiple for display.

For example, a 64 × 64 result exported at 8× is a 512 × 512 PNG. Each working pixel becomes an 8 × 8 block. With Keep aspect ratio, the result may be rectangular, so check the dimensions shown below the preview.

Work out where the softness comes from

The source may already be soft. Out-of-focus photos, JPEG artefacts and antialiased lettering bring blended edges into the conversion. Nearest-neighbour sampling cannot turn those into deliberately drawn lines.

The small grid may not have enough room. If a feature disappears at 32 pixels, try 64 or 128, or crop closer. An enlarged export will only make the existing small-grid pixels bigger.

The viewer may be smoothing the file. Compare the PNG at 100% and at a whole-number zoom. If blocks are sharp there but soft after posting, the receiving app may have resized or re-encoded it.

There is a tool-specific limit too: images above 1600 pixels on the longest edge first get a smoothed working-size reduction. The final pixel-grid step uses nearest neighbour, but the pipeline is not point-sampled from every full-resolution original. Use a well-prepared crop when fine source detail matters.

Use the palette to separate shapes

Try a small range of colour counts, such as 8, 16 and 32, while keeping the grid size fixed. Look at whether the subject separates from the background, not just whether the colours resemble the photograph.

The tool can extract a palette, use a built-in set or import your own hex, text, CSS or GIMP palette file. The palette extractor can help you prepare a set first. Its For pixel art export is a hex list ready to import.

A restricted palette is a useful stylistic choice, not a definition of all pixel art. Some images work with two colours; others need more than 32.

Add dithering when it helps the shading

Dithering trades flat bands for patterned mixtures of the available colours. Try Floyd–Steinberg on a gradient, Atkinson for a starker treatment or Bayer for a regular grid. The dithering guide explains the differences, and the dither tool offers a larger working image and halftone effects.

For an icon or a small face, the extra speckles may compete with the shape. Compare against None at the intended viewing size. Dithering can also increase PNG size by replacing flat areas with more varied patterns; compare the downloads if size matters.

Check transparency on the background you will use

With Keep transparency enabled, this converter makes each pixel either fully transparent or fully opaque, using a midpoint alpha threshold. That creates hard edges, but removes translucent shadows and effects.

Partial transparency is valid in pixel art; this particular tool does not preserve it. If it is part of your artwork, keep the original and use an editor that supports it. Inspect the result on both light and dark backgrounds for halos or unwanted edge colours.

Export for editing and for sharing

Use 1× when another editor or game engine should receive the actual pixel grid. Use 2×, 4×, 8× or 16× when you want larger blocks in the file itself. Pick a size close to the destination’s requirements rather than choosing the largest export automatically.

Whole-number enlargement keeps block widths consistent. A social platform can still rescale that image, so check the posted version. On a website you control, CSS image-rendering: pixelated can help preserve a pixelated appearance; integer display dimensions give the most predictable grid.

How the conversion works

Nearest-neighbour sampling copies a source pixel for each destination pixel rather than blending nearby samples. It can keep boundaries crisp, but can also miss thin features. Whole-number upscaling repeats the chosen pixels without inventing intermediate shades.

Palette reduction is a separate step. This tool uses a median-cut variant to choose automatic colours, then compares colours in Lab when matching pixels to the palette. Dithering changes how that matching error is spread.

The result can be a useful base to edit. Manual cleanup is where you can reshape an outline, enlarge an important feature or remove distracting pixels that a mechanical conversion kept.

Sources and further reading

• Heckbert, P. “Color Image Quantization for Frame Buffer Display.” SIGGRAPH ’82
• Floyd, R. W. & Steinberg, L. “An Adaptive Algorithm for Spatial Greyscale.” Proc. SID 17(2), 1976
• Bayer, B. E. “An optimum method for two-level rendition of continuous-tone pictures.” IEEE ICC, 1973
• CIE 15:2018 Colorimetry — CIELAB
• IEC 61966-2-1 — sRGB
• MDN — image-rendering and pixelated scaling

Last reviewed 2 October 2026.