Looks like it was this commit:
The “(HDR Menu) Brightness” setting in the appearance section is also broken now, but i don’t have enough builds to bisect if that was caused by the same commit.
Looks like it was this commit:
The “(HDR Menu) Brightness” setting in the appearance section is also broken now, but i don’t have enough builds to bisect if that was caused by the same commit.
It is having zero effect on my HDR Guest Presets.
It’s doing nothing but should be making things brighter?
I don’t have any HDR-capable setups, but the guy who made the changes said it’s working fine for him in vulkan.
he asked me to share with you guys the roadmap he has planned for HDR stuff, if you all have any suggestions: https://github.com/user-attachments/files/31055976/libretro_hdr_metadata_implementation_plan_v2.2.pdf
EDIT: there’s also some stuff about color management in the pipeline. ACES 2.0, etc.
You know, there’s a saying, if it ain’t broke, don’t fix it.
The old way of doing things - relying on RTINGS to determine Peak Luminance was flawed, confusing and now behind a paywall.
So the move to a simplified WYSIWYG actually did make things a lot simpler and the quality was there too.
At least @MajorPainTheCactus liased and actively sought the opinions of stakeholders here before doing anything major and he too had his plans for HDR going forward. So it would be nice if your contact would consider all of that while executing their vision that there are many users of these things who they need to consider, especially when it comes to stability and backward compatibility.
I know that everything wasn’t working properly in the HDR world like screenshots and the like, so focus on those low hanging fruit, but don’t unilaterally impose your vision on a community without first seeking their opinions.
I still need to dig into the pdf more, but my preliminary understanding based primarily on the commit is that this new setting is intended to support hypothetical native HDR cores, so it shouldn’t effect the brightness of existing HDR shaders.
There’s another saying: you can’t make an omelette without breaking a few eggs.
Unfortunately, HDR is still very much a technology in flux, so changes/breakages will happen if we want to keep up with it.
Yes, HDR cores are part of the changes, and the Peak Brightness thing is currently only hooked up for cores. Prboom and Doom3 on Windows with OpenGL (oddly enough) and Beetle-psx-hw with vulkan are all internally HDR now, so there should be a lot less quantization happening throughout the render process. I haven’t used it myself, but I assume this will make a big difference in the notoriously dark/can’t-see-shit Doom3.
I read this:
“libretro HDR Metadata Consolidation Implementation Plan for RetroArch and the libretro API Consolidating experimental HDR environment callbacks into a single structured query + optional inverse-tonemap operator modes + related driver-side HDR metadata gaps + guidance derived from advanced HDR shader development (Sony Megatron et al.) Status: Proposal (Revised) | Date: 2026-08-14 | Scope: libretro.h API + RetroArch frontend (Vulkan / D3D11 / D3D12 / GL1 / GL2 / GLCore / Metal) 1. Motivation and Current State The libretro API currently exposes five separate experimental environment callbacks for HDR presentation state: • RETRO_ENVIRONMENT_GET_SCREEN_10BPC_CAPABLE (88) • RETRO_ENVIRONMENT_GET_HDR_PAPER_WHITE_NITS (89) • RETRO_ENVIRONMENT_GET_HDR_EXPAND_GAMUT (90) • RETRO_ENVIRONMENT_GET_HDR_OUTPUT_MODE (91) • RETRO_ENVIRONMENT_GET_HDR_MAX_NITS (92) These values are tightly coupled. A core or shader emitting RETRO_PIXEL_FORMAT_HDR10_2101010 (or the equivalent slang #pragma format A2B10G10R10_UNORM_PACK32) almost always needs several of them at once when rebuilding colour tables or deciding ownership of encoding. Separate calls risk inconsistent snapshots if the user changes settings between queries, and they burn environment command IDs for what is logically one presentation contract. Key fact: As of August 2026 these callbacks are only used by RetroArch-internal code. No external cores have shipped against them. This remains the ideal window to consolidate and remove the individual experimental entry points without breaking the public ecosystem. However, the practical consumers of the HDR presentation contract are no longer limited to internal code. The multi-year development of high-fidelity HDR CRT shaders (most notably the Sony Megatron Colour Video Monitor family and its derivatives) has made the need for a clear, versioned, and reliable contract acute. Shader authors repeatedly encounter: • Double application of inverse-tonemap / paper-white scaling when a shader already owns absolute PQ luminance. • Confusion between RetroArch menu HDR settings and per-shader parameters (users and authors both). • Lack of a single authoritative snapshot of output mode, expand-gamut policy, and whether the frontend will leave the frame alone. • Linux Vulkan presentations that ignore static HDR10 metadata, producing over-bright / incorrectly scaled results (issue #18186). • UI / menu content that shares an incorrect tonemap path with game frames when HDR is active. The consolidation proposed here directly addresses the first three points and provides the foundation needed to make the remaining driver-side and UI issues solvable without further API churn.”
It sounds like it was written by AI. The main problem is a lack of consultation with the people who actually work closely with HDR on RetroArch. MajorPainTheCactus already did a major rewrite of RetroArch’s HDR pipeline earlier this year and addressed some of those same things stated in the design goals. Why the overlap? What’s wrong with @MajorPainTheCactus’ work? He definitely consulted us here. Who is this mystery developer? Why aren’t they commenting here? Did they even read what was done with HDR this year and what was still needed to be done? There are changelogs right here in this thread.
There’s something about the way this has been done and is being done that’s leaving a bad taste in my mouth. Can you ask your user not to interfere with the hard work that @MajorPainTheCactus put in this year to improve HDR in RetroArch please?
I’d rather talk about whether the proposals are right — that’s the only thing that matters here. It’s a working technical document, an RFC and a roadmap, and it’ll keep being updated as feedback comes in. That’s what I posted it for.
On the bigger concern, I think there’s a misunderstanding worth clearing up. Publishing this and asking hunterk to post it in this thread is the consultation. The alternative — the one I deliberately didn’t take — is pushing straight to a branch or master and letting everyone find out afterwards. You’re the audience I wanted eyes on this from.
None of this is aimed at undoing @MajorPainTheCactus’ work — his rewrite was a big step forward and it’s the foundation the roadmap builds on. The overlap in the design goals is there because those goals aren’t finished yet. HDR is still broken in a lot of places: Android, Wayland missing the metadata extension, and so on. There’s information a core currently has no way of obtaining at all — maxfall, maxcll — plus more operators than we expose today, and potentially colour management on top. I could write a book about the gaps.
Worth saying plainly: most of what’s in that PDF won’t affect shader authors at all. The parts that would touch you, like PASSTHROUGH, are the ones designed to help — and nobody has commented on those yet. Most of the rest is about accommodating true HDR cores and making the frontend play ball with them. If it lands well, a preset maker in your position shouldn’t notice much of it.
There’s something about the way this has been done and is being done that’s leaving a bad taste in my mouth. Can you ask your user not to interfere with the hard work that @MajorPainTheCactus put in this year to improve HDR in RetroArch please?
I’d rather have the technical conversation than that one. “Don’t change anything” isn’t something I can act on, but specific objections are — if there’s a proposal in there you think is wrong, or a regression risk I’ve missed, tell me which one and why. Genuinely, that’s the most useful thing you could do right now.
Timeline so nobody’s caught out: I move to implementation on 1 September, targeting mid-September for the work itself. Whatever feedback lands before then shapes what gets built. Happy to walk through any section of the document in more detail if that helps.
To be fair, I’d have to read the entire pdf document before I comment on any of that stuff but I know that MajorPainTheCactus had already implemented mechanisms for users who wanted the legacy behaviour - which is the Peak and Paper White Luminance settings being exposed in the shader and overriding the HDR Menu’s settings. To do this one has to append the hdr_config.slang file to any shader / shader chain.
He also made another slang available, which would add inverse tonemapping to any shader but not expose the Peak and Paper White Luminance settings and uses the HDR Menu’s settings for HDR Brightness.
Do what you need to do then, I trust that you would have already read the feedback and changelog in this thread during the HDR pipeline upgrade to be able to understand MajorPainTheCactus’ vision and the views of others’ here which were taken into consideration before arriving where we are or were before the recent changes.
I understand now that the scope of this is much wider than where things were and even more stuff might have to be changed in order to accomadate. The one thing that probably freaked me out the most and made me think that MajorPainTheCactus’ work was being reversed was the re-addition of the Peak Luminance setting in the HDR Menu after spending the last few months getting used to it’s simplicity and enjoying it. It’s not really that simple for users to calibrate things using separate Peak and Paper White Luminance values when there is no real point of reference for those values and the fact that they aren’t absolute since they can be easily skewed by things like Gamma adjustments and the like, not to mention that Peak Luminance varies by content and measurement window size.
All of the things would have contributed to use confusion and in one fell swoop, HDR Brightness solved all of that. All users have to do is load there preset and if it’s tpp dark, increase the HDR Brightness, period.
So when I see talk about user confusion and stuff like that still being mentioned in your document, it lead me to wonder if you would have been party to all that went on here within the past few months.
So please forgive my tone if I came across a little strongly. It is not that I am against change and progress. In addition to that, I am about to release my first shader project so it is imperative that things are relativey stable on the HDR front and I thought that we had reached to that point with respect to the vast majority of the Settings Menu options.
It would have been nice if one of the builds before these new changes to the HDR Menu and system could have been declared a stable build and given a new version number.
That would make it so much easier than keeping track of which nightly worked and which nightly broke this or that. It’s been too long and users need a milestone build and feature freeze before moving things too far forward in my opinion. What buil am I supposed to recommend my users who need to use the new HDR stuff? I can’t recommend the last stable so I have to recommend a nightly, but which one exactly? So you see my delimma.
I’m currently using one from around 15/04/2026.
Even this build as well as the latest nightly leave stuff to be desired as many of my Mega Bezel presets which have very long filenames and paths don’t load anymore. There seems to be some new character limit being enforced which was lower than what was there before. I do appreciate the multi-threaded shader loading though.
So yes, I do or at least did see the reimplementation of Peak Luminance in the HDR Menu as a regression.
Interesting, can you report this in a Github issue in more detail? It can be dealt with then
It really was just a stopgap for cores until we have proper metadata reporting in. Don’t be surprised if it goes away again. That is why time is crucial for the HDR implementation plan. The longer we wait finalizing the new spec, the more you run the risk that old habits will die hard in the future if people start designing their cores around what currently exists for HDR. Right now there is still the window of opportunity to nail it down correctly once and for all and then be able to consider it done for the next few years as far as the core-to-frontend information exchange goes
A post was merged into an existing topic: CD Audio cuts out during attract mode in Shadow of the Beast
Hello, I’ve been trying out this shader on my SDR screen, I’m really impressed with how it alters colors, I’ve never seen another shader that comes as close to the color enhancement of CRTs. Thank you
i’ve tested dozens of shaders and this one consistently looks the best especially with HDR enabled, thank you for your work.

Hi @MajorPainTheCactus, thank you for Sony Megatron. It’s the best CRT/HDR look I’ve seen.
I’m making an iOS/iPadOS video player app with real-time CRT shaders in Metal, and I’ve hand-ported Megatron v2 to it (including the HDR output). I’d love to include it in the public release, with full credit to you in the app and its licence notices.
Since the repo has no licence file, I assume all rights are yours, so I’m asking first: would you allow me to include my Metal port of Megatron v2 in the app, with credit? I’ll gladly take it out if you ever ask.
If there’s a better way to contact you, I’m happy to move this elsewhere. Thank you!
Megatron v2 + HDR10: double processing when RetroArch’s HDR Scanlines option is enabled
Setup:
Issue: With RetroArch’s HDR Scanlines option enabled (Settings > Video > HDR), Megatron v2 in HDR10 output mode looks heavily oversaturated and far too bright, even with Colour Boost set to Accurate. In scRGB mode the same setup looks correct.
What I found:
Additional note: Toggling this option requires a RetroArch restart to take effect, but the setting’s label does not mention this. I initially assumed the option did nothing, which made the issue hard to track down.
Suggestions:
Thank you very much for all the work you’ve done for this shader and also the team of RetroArch over these past years, you’re seriously amazing !
Would you mind running these tests on your setup?
The above should do.
Also, do be mindful that the new CRT-Guest-Advanced, Adamant as well as the latest CyberLab Guest Legendary and Golden Bezel contain additional subpixel modes which are supposed to support the new 4 Stack Tandem RGB layout of the LG G5 and some newer OLED TVs.
I went ahead and read the proposal. I think it’s making these arguments:
The document argues that the current API is incomplete, but most examples demonstrate missing platform implementations (Wayland, Android, HDR metadata reporting, display capability queries) rather than deficiencies in the core-to-frontend HDR interface itself. I don’t see how that’s an argument to change the interface. What are the pain points of implementing the missing platforms, and how do the proposed changes remove those pain points? I’m not seeing the connection. What engineering problem does moving callbacks to a struct solve?
As for a HDR contract and HDR ownership, I would say that’s already satisfied. Adding explicit indicators adds architectural complexity. Does it have a real benefit? The HDR contract is pretty clear: if the last shader has a 10-bit or 16-bit format, it’s HDR. For cores, ownership already appears to be implied by the existing HDR signaling and callback usage. The proposal should explain which ownership case cannot currently be expressed, i.e. what failure mode exists today that a passthrough indicator would resolve?
As for the peak brightness debate, I would challenge why RetroArch needs to have a built-in inverse tonemapper in the first place. RA should prefer delegating the work of SDR to HDR to the OS, and if that’s not possible, do a simple mapping based on a single target nits (e.g. 0-1 -> 0 to 100 nits). This is easy to understand, easy to validate, requires no ‘peak brightness’, and reduces scope of what frontends need to do. If HDR dedicated cores, shaders, etc. need additional parameters, have them define it as their own options. If someone wants to apply inverse tone mapping or any kind of arbitrary SDR to HDR conversion, dedicated shaders, maintained by people who care about them, can be used. In fact, MajorPainInTheCactus already does this, so he essentially has to implement his inverse tonemapping logic twice. I think it makes a lot of sense to relieve him from that obligation. What would make sense, for the front end to scope, is a setting for a default SDR -> HDR shader that applies to all cores/content that doesn’t have its own shader defined.
Can someone provide a concrete example of a core or shader that cannot be implemented correctly using the current HDR interface but would become possible with GET_HDR_METADATA ? I think that would go a long way to making it clear to everyone why the changes are needed.
I think this is an important distinction worth looking into. If all three phosphor channels have non-zero values in that arrangement, then it seems that all three simulated phosphors are contributing to the displayed red. The overall chromaticity may still be correct mathematically, but the actual subpixel behaviour is a separate question. It would be interesting to compare the individual channel output with a real CRT phosphor response, rather than relying only on the final colour appearance. That might help clarify whether the difference is purely visual or related to the underlying simulation.