HDR logos are a one-click export now. Keeping the gain map alive is the hard part.

Start with the arithmetic, since it is the only part of this you can verify without buying a screen. Two to the power of 2.9 is 7.46, so the multiplier the site quotes is internally consistent with the stop count it recommends ^[17][7]. Divide the other way and the implied reference white sits at about 201 nits ^[18]. That figure is the load-bearing one, because a display applies the multiplier against whatever it treats as white, and how much of the boost survives is a property of the panel and browser in front of the visitor, not of the file ^[2].

The delivery risk is structural rather than a bug in anyone's code. A gain map is a second grayscale representation carried inside or alongside the ordinary image, with metadata a decoder uses to work out how much brighter each pixel may go; a decoder that does not understand the metadata shows the base image ^[4]. Every transform between export and browser is a chance to lose that second payload: re-encode to WebP for a smaller response, or resize for a hi-dpi variant. Nothing throws an error. The mark renders in SDR, which is precisely what the fallback promised, so the loss never arrives as a ticket ^[19].

The tool's own restraint is worth noting against the temptation to boost everything. Whites are picked automatically, the user nominates additional brand colours, and the site says dark colours tend to come out as washed-out neon ^[6].

On live text, the costs are layout rather than bandwidth. Selection needs its own visible colour, overflowing lines can be clipped by the background technique, and the source notes that ordinary CSS layout choices can still expose edge cases even where rendering support has landed ^[13][14].

None of the underlying capability is new. The open-source gainmap-js project already encodes and decodes gain maps in JavaScript, and superwhite demonstrated brighter-than-white browser content using a tiny HDR video ^[15]. What changed is packaging: the encoding step disappears into a web form ^[1], and LinkedIn gets a separate BT.2100 PQ export with a matching ICC profile, which the site says the platform preserves ^[9].

Treat the brightness numbers as vendor figures. They come from the site, as does the claim that uploads are processed in memory and never stored ^[10], and the homepage does not establish when the product went live ^[16].

Which makes the decisive test a fetch rather than a look. Pull the production URL of a deployed mark and confirm the gain map is still in the delivered bytes after the pipeline has had its way. If it is, the effect is real for the visitors whose hardware can show it. If it is not, the asset is an ordinary JPEG that took extra steps to produce.