Fix Area Light Specular Over-Brightness

[aevyrie]

Objective

• Fix the over bright specular in our area (point) light implementation.
• Close Specular from area lights is disproportionately bright #13327

Solution

• Apply the modified (a_prime) specular roughness instead of the base roughness. This allows specular to correctly broaden and fade with distance per Karis 2013.
• The only downside of this approach is that specular reflections on smooth materials are over-attenuated. This is a known issue of Karis 2013, and the solution is to implement a popular solution like Linearly Transformed Cosines to handle area lights more correctly.
• In the meantime, I've applied a correction that lerps between the base roughness and the modified area light roughness, to retain sharpness and brightness for glossy materials. This was tuned from the reference renders.

Testing

Reference (Blender)

Before

After

Sharp Reference (Blender)

Sharp Before

Sharp After

---

Showcase

[github-actions]

Your PR caused a change in the graphical output of an example or rendering test. This might be intentional, but it could also mean that something broke!
You can review it at https://pixel-eagle.com/project/B04F67C0-C054-4A6F-92EC-F599FEC2FD1D?filter=PR-22372

If it's expected, please add the M-Deliberate-Rendering-Change label.

If this change seems unrelated to your PR, you can consider updating your PR to target the latest main branch, either by rebasing or merging main into it.

[aevyrie]

Before

Screen.Recording.2026-01-04.at.2.11.09.AM.mov

After

Screen.Recording.2026-01-04.at.2.11.44.AM.mov

[alice-i-cecile]

Yay! Progress towards finally fixing our specular highlights is huge. I've marked this as Controversial because I want SME-Rendering to decide whether we ship this as a stop-gap, or wait for a better fix (probably Linear Transformed Cosines) to avoid altering the appearance of this twice for our users.

[alice-i-cecile]

Per discussion in the Discord thread I opened to discuss this, consensus seems to be that is safest to ship in 0.19, but a last minute 0.18 fix would be tolerable. Leaving out of the 0.18 milestone, but @cart or @mockersf can overrule me if they feel strongly.

The assorted experts feel strongly that we should ship this, and not wait for LTC to further improve things.

[pcwalton]

Can you link me Karis 2013 (ideally the part of it where it says to do what you're doing)?

Also it'd probably be good to mention that linearly transformed cosines is a better solution in the comment you left, preferably with a link so that people who want to contribute the right solution know where to look.

[janhohenheim]

@pcwalton looks like page 14 of https://cdn2.unrealengine.com/Resources/files/2013SiggraphPresentationsNotes-26915738.pdf

This is the reason why the commented out code is... commented out.

The pages after that deal with the sphere light formula used here:

with the exception of the custom modification cc_lerp = 1.0 - (1 - a)^4.

[JMS55]

See also https://seblagarde.wordpress.com/wp-content/uploads/2015/07/course_notes_moving_frostbite_to_pbr_v32.pdf section 4.7.2.2 @aevyrie

[janhohenheim]

I don't have thaaaaat much knowledge here, but from comparing it with the paper it looks correct, and the manual tweak on top looks good in your test scenes :)

[pcwalton]

Awesome, let's get this in. Big improvement that addresses a long-standing problem.

[alice-i-cecile]

I'm leaning toward landing this in 0.18 as well. The current visuals are so clearly broken and so clearly a "quality perception liability", both for Bevy games and for Bevy generally, that any immediate forward progress would be welcome. But this (partially) has literature to back up the changes / seems well motivated.

Re-triaging to 0.18 per @cart's latest comments :) YOLO, 🚀, etc. etc.