"Convert JPG to PNG with a transparent background" is one of the most common image requests, and it hides a misunderstanding. A conversion changes how pixels are stored. It cannot invent information that was never in the file, and a JPEG never contains transparency.

That does not make the goal impossible. Removing a background is a real operation, and it is best done after the format conversion rather than during it. This guide separates the two steps and covers the part that decides whether the result looks professional: the edges.

01Why a JPG cannot become transparent

PNG and JPEG store different things. The W3C PNG specification describes the format as supporting indexed, greyscale and truecolour images "plus an optional alpha channel". That alpha channel is the per-pixel opacity value that makes part of an image see-through.

JPEG has no equivalent. It was designed for photographs, where every pixel is opaque, and the standard image model has no alpha component to store. So the information simply is not there: a JPG pixel is a colour with no opacity attached.

Converting the container cannot add it back. When a JPG is decoded, the pixels are colours; when a PNG is encoded, those same colours are written out with full opacity. The result is a valid PNG that looks exactly like the original, because it is exactly the original.

Format conversion preserves pixels; it does not edit them.

JPG to PNG gives you lossless storage of the same opaque image. Transparency requires a second step that decides, for every pixel, whether it should stay visible. That step is an edit, not a conversion.

02What the converter actually does

The UrsaKit JPG to PNG converter is the format step, and it is worth doing deliberately.

1. Drop the JPG onto the page, or click to browse for it.
2. Leave the output format on PNG.
3. Press the convert button and wait for the batch to finish.
4. Download the PNG. It is lossless, so repeated editing will not add another generation of compression artefacts.

The output keeps every pixel opaque. That is the correct behaviour, and it is what you want as the input to a background removal step: you now have a copy of the image that will not degrade further from re-encoding.

Two details are worth knowing. First, PNG is lossless, so the file usually grows for photographic content; in a local test with Pillow 11.3.0 on 2026-10-04, a 1200x800 photo-like image went from 81,069 bytes as JPEG at quality 85 to 1,160,327 bytes as PNG, about 14.3 times larger. Second, the reverse direction is not symmetric: the site's PNG to JPG converter fills transparent pixels with white before encoding, because JPEG has nowhere to put them.

03The three ways to get a transparent PNG

Once you accept that the background has to be removed rather than converted, there are three practical approaches, and they suit different images.

The colour key is the only one you can do with no model and no artwork, and it is also the one that interacts worst with JPEG. That interaction is worth understanding before you blame the tool.

04What JPEG does to edges

JPEG is lossy, and it stores the image in blocks of 8 by 8 pixels. A hard edge gets smeared across those blocks, so pixels that were pure background in the original end up slightly different from the key colour you are trying to remove.

To measure it, a 400x400 test image was created with a pure white background and a blue circle, then saved as JPEG at quality 85 and reopened. Of the 134,555 background pixels outside the circle, 891 deviated from white at all, 299 deviated by more than 2 levels, 34 by more than 8, and the largest deviation was 19.

background pixels sampled            134,555
deviation > 0                          891
deviation > 2                          299
deviation > 8                           34
deviation > 32                           0
max deviation                           19

That is why a naive colour key leaves a grey or coloured halo around the subject. A tolerance large enough to remove those 891 pixels also starts removing pixels that belong to the subject. Feathering the mask by a pixel or two, or shrinking it slightly after keying, hides the seam better than raising the tolerance.

Work on the largest source you have, and do not re-save as JPEG in between.

Every JPEG save adds another generation of ringing around the edges. If you must round-trip, keep the intermediate as PNG or TIFF from the first conversion until the final export.

05A workflow that produces a clean cutout

The order of operations matters more than the specific editor. This sequence works with a browser converter plus any tool that can key out a colour.

1. Convert the JPG to PNG first, so later saves do not add another JPEG generation. Use JPG to PNG for that step.
2. Identify the background colour by sampling a pixel well away from the subject, not next to it.
3. Set a tolerance just wide enough to clear the noise you measured. Start small and raise it until the halo disappears, then stop.
4. Check the subject's edge at 400% zoom. If it is chewed, lower the tolerance and feather the mask instead.
5. Inspect the alpha channel directly, not the preview on a white canvas. A checkerboard or a dark background reveals stray opaque pixels that white hides.
6. Export as PNG, then run it through an image compressor if the file is too heavy for its destination.

For icons and illustrations, the last step is often PNG to WebP rather than compression. WebP keeps the alpha channel and typically compresses flat graphics far better than PNG, at the cost of needing a fallback for very old software.

06What transparency costs in file size

A transparent PNG is still a PNG, and PNG is lossless. That is the trade: exact pixels and a real alpha channel, paid for with bytes.

The measured factor depends entirely on the content. A flat icon on a transparent background may shrink or barely change, because large uniform areas compress well. A photograph with a cut-out subject can grow by an order of magnitude, because every encoded pixel carries colour and opacity data that JPEG never had to store.

If the transparent asset is going on a website, the decision is usually not PNG versus nothing, but PNG versus WebP. Use PNG for the master copy you edit, and WebP for the copy you ship. Keep the master in a lossless format so a future colour or edge change does not start from a compressed source.

07Common mistakes

08Sources

• W3C — PNG Specification (Third Edition): optional alpha channel and image types (accessed 2026-10-04)
• MDN — Image file type and format guide: format capabilities including transparency (accessed 2026-10-04)
• Pillow documentation — the imaging library used for the size and edge measurements in this article (accessed 2026-10-04)