What is an HDR photo?

An HDR photo, as phones save it, is an ordinary photo plus a gain map — a second image inside the same file that records, for every pixel, how many times brighter that pixel may be shown on a screen that can go beyond standard white. An HDR display applies the map as far as its own headroom allows; any other screen ignores it and shows the ordinary photo. It is made for display, not a camera-raw or EXR-style HDR master. Below, a model predicts a gain map for your own photo by estimating how much brighter its highlights and light sources were than an SDR file can store.

  1. Your photo

    Drop a photo here — or click to choose

    JPGPNGWEBP

    Your files are never uploaded — everything runs on your device.

  2. HDR model

  3. Adjust & export

Try it on one of your own photos

  1. Add your photo

    Drop a JPG, PNG or WebP. It loads straight into your browser — your files are never uploaded.

  2. Let the model predict the gain map

    GMNet looks at your photo and predicts how much brighter each part of it should be on an HDR screen. The first run downloads an ~8 MB model, cached for next time.

  3. Adjust and download the Ultra HDR JPEG

    Turn the predicted boost down or up with a 0.5×–1.5× strength setting, then download one JPEG (up to 4096 px on the long side): your photo, re-saved as the base image, with the gain map embedded. HDR screens apply it; other screens show the base.

What HDR does, and what it doesn't

It changes brightness, not detail

HDR on a photo is about how bright a highlight is allowed to get on your screen — a specular reflection, a lamp, sunlight on water. It is not extra sharpness and not extra colour information hiding in the file.

One file, two screens

An Ultra HDR JPEG is a standard JPEG with a second JPEG — the gain map — embedded in it, plus metadata that sets the map's range. Each gain-map pixel encodes, on a log2 scale, the brightness ratio between the HDR and SDR versions. An HDR display applies that boost in linear light, scaled back if it has less headroom than the map asks for; an SDR display ignores the map and shows the base.

It cannot bring back what was clipped

A normal photo stores nothing brighter than white — anything brighter was capped at white. The model can estimate how much brighter a blown-out lamp or sky should glow, but not the detail inside it: a clipped highlight comes back brighter, not more detailed.

The model behind it

Powered by GMNet (ICLR 2025, MIT licence), which predicts the gain map from the photo itself, so the boost follows the content rather than a fixed brightness curve. The boost tops out at 5× the SDR brightness (about 2.3 stops), even at the strongest setting, and the file is an Ultra HDR JPEG — Google's gain-map format for Android, using Adobe's gain-map metadata.

Frequently asked questions

What is an HDR photo?
It is a photo whose highlights may be shown brighter than standard white on a screen that supports it. In Ultra HDR that instruction travels as a gain map inside an otherwise normal JPEG, so the same file looks right on an HDR screen and on an ordinary one. It is not the same as a camera-raw file (the sensor's data) or an EXR (scene brightness as floating-point values), both of which are made for editing rather than display.
What does HDR do on a photo?
On an HDR screen, the bright parts — reflections, lights, sky — are shown brighter than the white of the interface around them, while darker areas mostly stay where they were. How much brighter depends on the screen's headroom. A standard screen ignores the gain map: you see the base image, which is your photo.
Why does my HDR photo look the same as the original?
Almost certainly because the screen you are looking at is not an HDR one, or the app you opened it in does not read gain maps. Open the same file on an HDR phone or monitor, in an app that supports Ultra HDR, and the highlights will visibly brighten.
Does this recover blown-out detail?
No. The gain map makes clipped areas glow brighter on HDR displays, but the detail lost inside them cannot be restored — an HDR look, not detail recovery.
Are my photos uploaded anywhere?
No. Everything — decoding, gain-map prediction and assembling the final JPEG — runs entirely in your browser on your own device. Your photo is never uploaded, and it is not kept once you close the tab.
Which model does it use?
GMNet, in the variant trained on real phone HDR photos. It predicts a 512×512 gain map that is scaled up to your photo's full size before saving, and runs on the GPU via WebGPU where your browser supports it, otherwise on the CPU.

Wondering what the difference actually is, or why HDR looks washed out on your monitor? HDR vs SDR — the one difference behind the fifteen bullet points, what the formats mean, and why every comparison image online is necessarily fake.