You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Pixel 8 Pro: the screen looked washed out after boot and deeper later, and I can't reproduce what changed it #537
This is a question, not a finding. On my 8 Pro the screen looked washed out from boot (12:24) until about 17:51 on 2026-11-08. After two power-key presses (lock, then wake) it looked deeper: darker tones and more contrast, a bit too dark, but I like it. I rebooted at 18:39 and it came back washed out, so it's not a one-time thing, but I haven't been able to make it flip again.
The picture is a pair of frames from a phone video, same camera and phone position, exposure lock not confirmed (a dark corner of the room reads 0.33 and 0.31 on a 0 to 100 scale in the two frames). Left is before the reboot (18:39:03), right is after boot (18:40:24). Screenshots can't show it: they have identical pixels before and after.
What I tried after the reboot, each one by eye, one change at a time, and none flipped it: the backlight at 0 and back, the whole range from 1 to the maximum 3307 and three fast drags, my real slider, a panel sleep and wake (kernel log 60 Hz then 120 Hz), the same with hyprsunset running, the real power-key pair, about 30 rapid power-key presses, and plugging and unplugging the charger. Before the reboot I also tried hyprsunset on and off, its temperature and identity, and Hyprland's sdr brightness and saturation: no visible change either. The battery current at the same backlight and the same home screen was 783.8 mA washed and 785.0 mA deep (median of 30 samples; the deep run had a short moment with the app drawer open), so no measurable power difference:
What the kernel source shows, from a read-through and not a line-by-line audit: gs-decon has no gamma, color matrix or dither, and the panel driver (hk3) sends brightness and the cold-init commands only and leaves HBM and IRC as the bootloader set them. gs_decon is loaded with screen_off=2 (from omamobile-device-husky), so each screen-off really sleeps the panel and the next wake restarts it. I can't tell what the bootloader leaves behind, and I couldn't read panel registers.
One more thing I checked: the night light and any color change through hyprsunset can't work on this display, which is a separate fact: Hyprland logs no ctm prop support and crtc doesn't support CTM for DSI-1, and the DRM state has no CTM or gamma property. So the colour layer isn't what changed here.
A logger now runs on the phone (every 10 s: backlight, panel state, refresh, SoC temperature, battery, plus power-key and display events) so I have the state at the moment it flips next time.
Question for anyone with a husky, a 7 Pro or a 9a: do you see a look that changes without any setting, and what did you press just before? Is it known that the first boot state differs from the driver's own setup?
An update on this, because I think I found what the washed look is, or at least a way to fix it on my unit. @derekross@greenbrettmichael, I'd like your eyes on it, and @gitColinX, yours too if your 8 Pro looks the same as mine.
What I tested, all on my 8 Pro with the screen on, by eye, blind where I could:
A full-screen black image looks the same as the screen switched off (4 rounds). So the panel's black is fine.
Hyprland's own transfer curve, render:cm_sdr_eotf default against gamma22: I couldn't tell them apart on a photo, blind (5 rounds with one identical pair, I said "the same" every time). So I can't say that setting does anything here.
A custom Hyprland screen shader that only does pow(rgb, g). Two blind rounds, names hidden, random order. In the first, 1.5 looked best and the plain picture, with no change, looked washed out. In the second, 1.6 looked best, 1.7 a bit dark, 1.35 a bit bright, and adding saturation to 1.6 was the worst. An earlier pass with the names showing also picked 1.5, but liked two values the blind pass rejected, so I trust the blind ones.
At 1.55 I loved the blacks but the bright colors looked too bright to me. A fourth round, blind again, tried a brightness gain on top of 1.55 (colors times 0.78 to 0.92), alone and with lower saturation. Gains of 0.85 and 0.92 looked best; the worst was 0.90 together with saturation 0.90. I run 0.88 now, between the two best, which wasn't a step in the test.
Screenshots and VNC can't show any of this: a 12-step gray ramp gave identical pixel values with every cm_sdr_eotf setting, so it has to be judged on the panel.
Why I think it is the gamma. Google's hk3 panel driver (panel-gs-hk3.c) has no color-mode or gamma table that I could find, apart from one tweak for a pre-production panel. Our display controller driver (gs-decon) has no gamma, color matrix or dither. I am guessing stock Android does that in the SoC, because it isn't in the panel driver. That would fit a flat, washed look that a gamma of about 1.5 puts right. It is from reading source: I could not read the panel's registers and the bootloader may still be leaving its own settings, it is one unit and my eyes, and it does not explain the one deep look I had on 11-08 or a shutdown logo that looked deep once. Our hk3 driver also doesn't send some commands Google's does (0x53 and the IRC sequence); I haven't tested whether those matter.
The shader runs on the GPU (Mali through Panfrost). While it is on, the CPU stays mostly idle and the GPU at its lowest clock, so it works as a stopgap. The real fix would probably be a gamma LUT in the display controller for zuma. Two questions:
Does your 8 Pro look washed out after boot too, and does a shader like this make it look right to you?
Would you take a husky-only stopgap in omamobile-device-husky (the shader plus two lines of config) until gs-decon gets a real gamma path, or would you rather wait for the kernel side?
I can put the shader, the numbers and the test scripts in a PR if you say where you want them.
A correction to my last comment here. I wrote that I'm guessing stock Android does the color work in a gamma table our kernel lacks, and that the real fix would probably be a gamma LUT in the display controller. I've since read the real stock calibration data (the same files are in the GrapheneOS image, and the panel driver in GrapheneOS's public kernel repo matches Google's file; I haven't checked that it is the branch of my build). The SoC part of my guess is right, but what it does there is gamut mapping, not a tone curve: the native regamma table is a plain 1/2.2 encode, and the color LUTs leave grays alone and only pull the panel's primaries, which are wider than P3, in. If the degamma is the usual 2.2 or sRGB decode (that table isn't in the files, the closed color library builds it at run time), stock's tone for mid-tones is close to identity: half in gives about half out. The shader I run gives 0.30 there, so a gamma table filled with stock's numbers would not give the curve that looks right in my blind tests. The blind tests stand, but the reason for the flat mid-tones is unknown, so please ignore my "gamma LUT" line.
What I can't exclude yet: the panel's own state without Google's init commands, a DQE or HDR stage the bootloader leaves on (its image has decon and hk3 strings but none for dqe, gamma or cgc, so it is probably off, but I haven't read the registers), or the default color mode. A DQE port would still matter for the gamut mapping; without it the colors are probably over-saturated.
@derekross, @greenbrettmichael: would you accept a read-only dump of the DQE control registers as the bootloader leaves them (it needs the region mapped, no /dev/mem sweep), and a read of the panel's registers through the DSI read path, if the published tree has it the way I read it? And does your 8 Pro look washed out after boot? That is still the thing I'd most like to know.
@Somnius, my unit doesn't look washed out to me. On the published kernel with gs-decon at 120 Hz, the picture looks clean and saturated, about like Omarchy on my laptop. It looked better again once 120 Hz was on. I haven't compared it side by side with stock, and I don't have a calibrated eye, so take it as one owner's impression. It may also be that panels differ between units.
@Somnius, a correction to what I said on 11-15. For a while that day my 8 Pro looked much deeper than I have seen it before or since: darker tones, more contrast. So my "clean and saturated" was probably that state. Since then it looks the way it always has, which is flatter, and neither switching themes nor the brightness slider brings the 11-15 look back. I can't say whether that normal look counts as washed out without a stock phone next to it.
Nothing changed in software around it: kernel 82f31d3a7, no package changes after 11-13, and hyprctl monitors reads DSI-1 1344x2992 at 120 Hz, XRGB8888, sRGB, sdr brightness 1, sdr saturation 1. The phone has restarted at least three times since (twice after running flat). I have no photo and can't say what flipped it, so take it as one sighting.
You wrote on 11-14 that you had solved the washed-out screen. What was it?
@gitColinX I posted the tests and numbers here on 11-14: #537 (comment), the correction after it: #537 (comment), and the "solved" note: #537 (comment). There's no PR for it yet, it only lives in my own config. It's now a husky-only PR: #880 (it builds on the device hook in #204).
In short, I didn't fix the cause, I made it look right on my unit with a Hyprland screen shader: pow(rgb, 1.55) then times 0.88. That came out of four blind rounds by eye on the panel (gamma 1.5 to 1.6 won every time, no change looked washed, more saturation always looked worse, and the 0.88 gain because 1.55 alone made bright colours too bright). Screenshots can't show any of it, it has to be judged on the panel. It costs almost nothing: the GPU stayed at its lowest clock with it on.
What it means: on my unit the mid-tones come out much flatter than they should, and the shader puts them back. Why is still unknown. As I wrote in the correction above, stock's own tables don't explain a curve like that, so the panel's own state after the bootloader is still the main suspect, and that would also fit your deeper look on 11-15 coming and going with nothing changed in software. I had the same once on 11-08 and on one shutdown logo.
Here is exactly what runs on mine. ~/.config/hypr/shaders/omamobile-look.frag:
It applies live. To undo, delete the line. If you try it, it would be good to know if 1.5 to 1.6 looks right on your unit too, or if yours sits somewhere else.
Activity
Somnius commentedon Nov 8, 2026
If @derekross and @greenbrettmichael have any ideas on this, please share them. The true oled blacks where amazing while on it...
Somnius commentedon Nov 14, 2026
An update on this, because I think I found what the washed look is, or at least a way to fix it on my unit. @derekross @greenbrettmichael, I'd like your eyes on it, and @gitColinX, yours too if your 8 Pro looks the same as mine.
What I tested, all on my 8 Pro with the screen on, by eye, blind where I could:
render:cm_sdr_eotfdefault against gamma22: I couldn't tell them apart on a photo, blind (5 rounds with one identical pair, I said "the same" every time). So I can't say that setting does anything here.pow(rgb, g). Two blind rounds, names hidden, random order. In the first, 1.5 looked best and the plain picture, with no change, looked washed out. In the second, 1.6 looked best, 1.7 a bit dark, 1.35 a bit bright, and adding saturation to 1.6 was the worst. An earlier pass with the names showing also picked 1.5, but liked two values the blind pass rejected, so I trust the blind ones.cm_sdr_eotfsetting, so it has to be judged on the panel.Why I think it is the gamma. Google's hk3 panel driver (
panel-gs-hk3.c) has no color-mode or gamma table that I could find, apart from one tweak for a pre-production panel. Our display controller driver (gs-decon) has no gamma, color matrix or dither. I am guessing stock Android does that in the SoC, because it isn't in the panel driver. That would fit a flat, washed look that a gamma of about 1.5 puts right. It is from reading source: I could not read the panel's registers and the bootloader may still be leaving its own settings, it is one unit and my eyes, and it does not explain the one deep look I had on 11-08 or a shutdown logo that looked deep once. Our hk3 driver also doesn't send some commands Google's does (0x53and the IRC sequence); I haven't tested whether those matter.The shader runs on the GPU (Mali through Panfrost). While it is on, the CPU stays mostly idle and the GPU at its lowest clock, so it works as a stopgap. The real fix would probably be a gamma LUT in the display controller for zuma. Two questions:
omamobile-device-husky(the shader plus two lines of config) untilgs-decongets a real gamma path, or would you rather wait for the kernel side?I can put the shader, the numbers and the test scripts in a PR if you say where you want them.
Somnius commentedon Nov 14, 2026
A correction to my last comment here. I wrote that I'm guessing stock Android does the color work in a gamma table our kernel lacks, and that the real fix would probably be a gamma LUT in the display controller. I've since read the real stock calibration data (the same files are in the GrapheneOS image, and the panel driver in GrapheneOS's public kernel repo matches Google's file; I haven't checked that it is the branch of my build). The SoC part of my guess is right, but what it does there is gamut mapping, not a tone curve: the native regamma table is a plain 1/2.2 encode, and the color LUTs leave grays alone and only pull the panel's primaries, which are wider than P3, in. If the degamma is the usual 2.2 or sRGB decode (that table isn't in the files, the closed color library builds it at run time), stock's tone for mid-tones is close to identity: half in gives about half out. The shader I run gives 0.30 there, so a gamma table filled with stock's numbers would not give the curve that looks right in my blind tests. The blind tests stand, but the reason for the flat mid-tones is unknown, so please ignore my "gamma LUT" line.
What I can't exclude yet: the panel's own state without Google's init commands, a DQE or HDR stage the bootloader leaves on (its image has decon and hk3 strings but none for dqe, gamma or cgc, so it is probably off, but I haven't read the registers), or the default color mode. A DQE port would still matter for the gamut mapping; without it the colors are probably over-saturated.
@derekross, @greenbrettmichael: would you accept a read-only dump of the DQE control registers as the bootloader leaves them (it needs the region mapped, no /dev/mem sweep), and a read of the panel's registers through the DSI read path, if the published tree has it the way I read it? And does your 8 Pro look washed out after boot? That is still the thing I'd most like to know.
Somnius commentedon Nov 14, 2026
@derekross, @greenbrettmichael I solved the problem with the washed out screen. I'll report as soon as I do more tests.
gitColinX commentedon Nov 15, 2026
@Somnius, my unit doesn't look washed out to me. On the published kernel with gs-decon at 120 Hz, the picture looks clean and saturated, about like Omarchy on my laptop. It looked better again once 120 Hz was on. I haven't compared it side by side with stock, and I don't have a calibrated eye, so take it as one owner's impression. It may also be that panels differ between units.
gitColinX commentedon Nov 18, 2026
@Somnius, a correction to what I said on 11-15. For a while that day my 8 Pro looked much deeper than I have seen it before or since: darker tones, more contrast. So my "clean and saturated" was probably that state. Since then it looks the way it always has, which is flatter, and neither switching themes nor the brightness slider brings the 11-15 look back. I can't say whether that normal look counts as washed out without a stock phone next to it.
Nothing changed in software around it: kernel
82f31d3a7, no package changes after 11-13, andhyprctl monitorsreads DSI-1 1344x2992 at 120 Hz, XRGB8888, sRGB, sdr brightness 1, sdr saturation 1. The phone has restarted at least three times since (twice after running flat). I have no photo and can't say what flipped it, so take it as one sighting.You wrote on 11-14 that you had solved the washed-out screen. What was it?
Somnius commentedon Nov 18, 2026
@gitColinX I posted the tests and numbers here on 11-14: #537 (comment), the correction after it: #537 (comment), and the "solved" note: #537 (comment). There's no PR for it yet, it only lives in my own config. It's now a husky-only PR: #880 (it builds on the device hook in #204).
In short, I didn't fix the cause, I made it look right on my unit with a Hyprland screen shader:
pow(rgb, 1.55)then times 0.88. That came out of four blind rounds by eye on the panel (gamma 1.5 to 1.6 won every time, no change looked washed, more saturation always looked worse, and the 0.88 gain because 1.55 alone made bright colours too bright). Screenshots can't show any of it, it has to be judged on the panel. It costs almost nothing: the GPU stayed at its lowest clock with it on.What it means: on my unit the mid-tones come out much flatter than they should, and the shader puts them back. Why is still unknown. As I wrote in the correction above, stock's own tables don't explain a curve like that, so the panel's own state after the bootloader is still the main suspect, and that would also fit your deeper look on 11-15 coming and going with nothing changed in software. I had the same once on 11-08 and on one shutdown logo.
Here is exactly what runs on mine.
~/.config/hypr/shaders/omamobile-look.frag:and one line at the end of
~/.config/hypr/looknfeel.lua(Hyprland 0.57):It applies live. To undo, delete the line. If you try it, it would be good to know if 1.5 to 1.6 looks right on your unit too, or if yours sits somewhere else.
omamobile-device-husky 0.1.17: screen look stopgap (derekross#537)