Field test · NiCE Cognigy
Not a glow filter. Flat images whose white pixels are encoded on the SMPTE ST 2084 curve at four to twenty times the brightness your screen renders ordinary white. Served here untouched, so all three encodings work at once.
All three put the same pixels on screen. They differ in how the file tells the browser those pixels are HDR, and therefore in what survives when a platform re-encodes the upload.
An ICC profile named “Rec2020 Gamut with PQ Transfer” rides along in the JPEG. This is what Wiz and Port.io shipped on LinkedIn.
Survives LinkedIn Page logos · no SDR fallback
Four bytes declaring BT.2020 primaries and PQ transfer, per PNG Third Edition. 16-bit, so no banding and no compression ringing.
Best for self-hosting · dropped on re-encode
A correct SDR image plus a per-pixel boost map. Scales to whatever headroom the display actually has, down to none.
Only one that looks right on SDR
Diffuse white is 203 cd/m² per ITU-R BT.2408. Everything above it is headroom. The scale is the PQ curve itself, not linear.
The confirmed route. Upload as the company Page logo and cover. Personal profile photos get re-encoded and lose it. Compare each mark against the white chip below it: if it does not out-shine that chip, the effect is not reaching you.
16-bit Rec.2020 PQ PNGs with a cICP chunk. This is the
right file for nice.com or cognigy.com, where nothing re-encodes it. Verify with
ffprobe -show_entries stream=color_transfer,color_primaries, which
should report smpte2084 and bt2020.
The hedge. On a display with no headroom these are pixel-correct sRGB rather than dark and smeared, which matters if a big share of your audience is on Windows work monitors.
2400 × 800 versions at several peak levels, plus a bloom variant and an inverted composition where the plate burns and the wordmark stays black. Each is paired with its SDR reference.
Drop in a PNG, JPEG or SVG and get a file you can upload straight to a
LinkedIn Page. Everything runs in this tab: the image is never uploaded anywhere.
The result appears below as a real <img>, so on an HDR display it
will glow right here.
Transparent PNG or SVG works best.
cd/m² for white. Wiz used ~4000; 1000 is restrained.
60 cd/m². Near-black has more contrast but collapses to a black box on SDR screens.
Survives a LinkedIn Page logo upload.
Page logo, Page cover, or leave as-is.
For a black logo. Neutral pixels flip; brand colours are left alone.
Paste this into Claude with your own logo attached.
I have a logo SVG and I want an HDR version where only the white/bright elements physically glow on HDR displays (MacBook Pro 2021+, iPhone 12+, modern Android). The technique: - Composite in ABSOLUTE linear light, in cd/m2, before any transfer curve. Antialiased edges must blend between a 1000-nit glyph and a dark plate in linear light, not in a gamma-encoded space. - Encode bright pixels on the BT.2100 PQ curve (SMPTE ST 2084) at 1000-4000 nits so they land in the display's EDR headroom. - Keep coloured pixels at the SDR reference of 203 cd/m2 (ITU-R BT.2408) so brand colours render at their true hue. Do the real sRGB -> Rec.2020 matrix conversion; do NOT just retag sRGB bytes as Rec.2020, that sends saturated colours to thousands of nits in the wrong hue. - Keep the background plate around 60 cd/m2 rather than near-black, so it stays readable on non-HDR screens. Produce three encodings and tell me which to use where: 1. JPEG with a Rec.2020 + PQ ICC profile embedded. This is what survives a LinkedIn company Page logo upload. 2. 16-bit PNG with a cICP chunk (9/16/0/1) and a cLLI chunk, for self-hosting. 3. An Ultra HDR gain-map JPEG, for graceful SDR fallback. Compute the gain map against the ENCODED base image, after clipping and JPEG compression, not against the ideal SDR array. Then verify your own output: decode the files back to absolute cd/m2 and show me the measured levels, and check ffprobe reports smpte2084 / bt2020 on the PNG.