Convert HEIC to JXL

An HDR photograph from a recent iPhone is not one image with a wider range of numbers in it. It is an ordinary standard-range picture plus a gain map — a low-resolution greyscale layer stored alongside it in the same container, describing how much extra brightness each area should get on a display capable of showing it. That is why the same file looks luminous in the Photos app and flat when something else opens it.

This conversion reads the container, takes the primary image and encodes that. The gain map sits beside it as an auxiliary picture and is not applied and not carried, so the JXL you get is the standard-range rendering. Nothing is corrupted and nothing is broken; the extra layer never enters the pipeline. If the glow in the highlights is why you like the photograph, this route will not preserve it and no setting on this page will change that.

JPEG XL is genuinely the most capable still format on this site. The specification, ISO/IEC 18181, allows up to 32 bits per channel, wide-gamut colour, alpha, animation and a lossless mode that is competitive with anything. Published as ISO/IEC 18181-1 in 2022 by the same committee that produced JPEG, it was designed with exactly the case you are here for in mind.

The pipeline that feeds it here is more modest. The HEIC is decoded to eight-bit RGBA, resized if you asked, and handed to the encoder at quality 82. A format that could hold four times the bit depth receives eight bits; a format that could carry the picture losslessly is asked for a lossy encode. That is not a fault to be fixed on this page, it is the shape of the conversion, and knowing it is the difference between a sensible decision and a disappointed one.

The property JPEG XL is best known for is that it can take an existing JPEG file and repack it, losslessly, at roughly a fifth less space, with the original JPEG bit-exactly recoverable afterwards. It is a remarkable piece of engineering and it is what people usually mean when they say JPEG XL is lossless.

It works because JPEG XL shares its ancestry with JPEG and can re-code the same coefficients. A HEIC holds HEVC-compressed data, which has nothing in common with that, so there is no repacking to do — the picture has to be decoded and compressed again from scratch. If somebody has told you that converting to JPEG XL is free of loss, they were describing the JPEG case and it does not transfer.

A HEIC can hold ten bits per channel and the iPhone uses that, most visibly in gradients: a clear sky, a wall in even light, skin in soft window light. Ten bits gives four times as many levels to describe those transitions as eight does, which is the difference between a smooth ramp and one with faint steps in it.

The decode here produces eight bits, so those levels are quantised before the encoder ever sees them. On a photograph with texture across the frame it makes no visible difference. On a big smooth gradient it can, particularly if the picture is later brightened or has its contrast pulled about — banding that was latent in eight bits becomes visible once the range is stretched.

iPhone photographs are tagged Display P3, a colour space noticeably wider than sRGB in saturated reds and greens. JPEG XL can carry a colour profile and does so well. This conversion writes none, so the numbers arrive unlabelled and any viewer will assume sRGB.

The result is a slight desaturation of the strongest colours rather than a dramatic shift — a red that was vivid comes back merely bright. It is the same behaviour as every other conversion on this site that decodes and re-encodes, and it is the second reason, after the gain map, that this route is not the way to move a colour-critical photograph. Editing the HEIC in software that manages colour and exporting from there is.

The encoder runs at quality 82 unless you change it, which is the same default the rest of the site uses and a sensible one: high enough that the second generation is not visible on ordinary subject matter, low enough that the file stays worth having. Raising it towards 95 costs a lot of bytes for very little; dropping it below 70 starts to show.

Where a second pass shows first is where the first pass already worked hardest. Look at out-of-focus backgrounds, at the boundary between a bright sky and a dark roofline, and at large areas of nearly-flat colour. If you are converting pictures you care about, convert five of your most difficult ones and compare at full size before committing to the rest.

Besides the gain map, a HEIC can hold a burst, the frames behind a Live Photo, a depth map from Portrait mode, or an unedited original kept beside the edited version. All of that is what makes the format interesting and all of it is auxiliary to the primary image.

The conversion takes the primary image, which is the one the phone shows and the one anybody pointing at the file means. Everything else stays in the HEIC. That is a strong argument for not deleting the originals after converting: the JXL is a rendering of one picture out of a file that contained rather more than one picture.

The full pixel dimensions come through unchanged unless you set a maximum width, so a 12-megapixel frame stays 4,032 by 3,024 and a 48-megapixel one stays 8,064 by 6,048. An alpha channel, if the source has one, survives the conversion — JPEG XL supports it and the pipeline preserves it.

The metadata does not. Capture date, camera model, exposure, lens and any GPS coordinate are absent from the output. For a photographer that is worth planning around rather than discovering: a folder of undated pictures is much harder to live with than a folder of large ones, so keep the HEIC originals or export the capture data separately before you rely on the JXL set.

Support for JPEG XL is partial and it has moved in both directions. Some browsers decode it and some have removed the capability; some image editors read it natively and others need a plug-in; a good many photo managers and upload forms do not know it exists. This is the single biggest practical risk of choosing the format.

The test is cheap. Convert one photograph, then try to open it in the specific things you use — your editor, your file manager’s preview, whatever gallery software the pictures will live in, and the browser you actually browse with. If all four handle it, the format is a good choice for you. If two of them do not, the conversion has produced files you cannot look at, and that is a worse outcome than a slightly larger format.

Both halves of this run locally: libheif decodes the HEIC and the JPEG XL encoder compresses the result, each compiled to WebAssembly and loaded on demand into the page you are reading. No request carries an image, which the network tab shows and a disconnected machine proves.

That is what makes experimenting practical here. Trying three quality settings on the same picture, or converting a batch and throwing it away, costs nothing but your own processor time — and it does not mean handing your photographs to an unknown server three times to find out which setting you preferred.