sky_line/Slipstream

Fork 1

RT shadow mask below full resolution: flickering dark texels on moving avatars (amplified by DLSS / FSR into 2x2 blocks) #1086

New issue

Open

opened 2026-10-25 21:56:25 +00:00 by sky_line · 1 comment

sky_line commented

2026-10-25 21:56:25 +00:00

Owner

Owner, 2026-10-25 (dev fc780b19b3, RTX 5090):

Small dark blocks scattered over avatars (hair, faces, horns, ears), changing every frame.
Only when RenderVKRTShadowMaskScale < 1 (the default is 0.5). Full resolution is clean.
Native: mostly gone, but small black pixels remain that change with the avatar's movement. DLSS amplifies them into visible dark blocks; FSR amplifies less.
Not affected by: GI on/off, RenderVKRTShadowRrHistory 0 (#1050), RenderVKReSTIRReuseMotion 0, RenderVKDenoiseMovingDepthTol 0, RenderVKDenoiseYoungBoost 0, RenderVKDenoiseDisoccFix 0, RenderVKLitDebugMask 8192. Stochastic local-light shadows on or off.

Likely area: the reduced-resolution mask's reconstruction to full resolution on moving / rigged geometry -- the depth/normal-aware upsample picking a texel from a different surface (background or an occluder) at avatar silhouettes and fine geometry (hair cards), the mask's temporal history / reprojection for skinned avatars, or the rigged-geometry trace at half-res sample positions (self-intersection on skinned meshes). Upscalers then lock the dark samples in temporally.

Acceptance: no dark texels on avatars at mask scale 0.5 under native, FSR and DLSS RR; static scenes unchanged.

Owner, 2026-10-25 (dev fc780b19b3, RTX 5090): - Small dark blocks scattered over avatars (hair, faces, horns, ears), changing every frame. - Only when `RenderVKRTShadowMaskScale` < 1 (the default is **0.5**). Full resolution is clean. - Native: mostly gone, but small black pixels remain that change with the avatar's movement. **DLSS amplifies them** into visible dark blocks; FSR amplifies less. - Not affected by: GI on/off, `RenderVKRTShadowRrHistory` 0 (#1050), `RenderVKReSTIRReuseMotion` 0, `RenderVKDenoiseMovingDepthTol` 0, `RenderVKDenoiseYoungBoost` 0, `RenderVKDenoiseDisoccFix` 0, `RenderVKLitDebugMask` 8192. Stochastic local-light shadows on or off. Likely area: the reduced-resolution mask's reconstruction to full resolution on moving / rigged geometry -- the depth/normal-aware upsample picking a texel from a different surface (background or an occluder) at avatar silhouettes and fine geometry (hair cards), the mask's temporal history / reprojection for skinned avatars, or the rigged-geometry trace at half-res sample positions (self-intersection on skinned meshes). Upscalers then lock the dark samples in temporally. Acceptance: no dark texels on avatars at mask scale 0.5 under native, FSR and DLSS RR; static scenes unchanged.

sky_line added the

rendering

bug

labels

2026-10-25 21:56:25 +00:00

sky_line added this to the Slipstream project

2026-10-25 21:56:25 +00:00

sky_line referenced this issue from a commit

2026-10-26 00:26:21 +00:00

#1086: RT shadow mask upsample -- thin receivers take the brightest tap close behind, not the closest-depth one

sky_line referenced this issue from a commit

2026-10-26 00:26:21 +00:00

#1086: A/B knob for the RT shadow mask upsample's close-range depth tolerance

sky_line referenced this issue from a commit

2026-10-26 00:30:04 +00:00

#1086: RT shadow mask upsample -- thin receivers take the brightest tap close behind, not the closest-depth one

sky_line referenced this issue from a commit

2026-10-26 00:30:04 +00:00

#1086: A/B knob for the RT shadow mask upsample's close-range depth tolerance

sky_line commented

2026-10-26 00:30:21 +00:00

Author

Owner

Owner test of the fix (c499b1a318 / cc6af06e47, on main and release/1.5): clearly improved under all settings; native looks clean. Some shadow noise remains under DLSS Ray Reconstruction and slightly under FSR. Keeping open for the remaining upscaler-amplified noise (candidates: the temporal pass's plane test still accepting scalp history for hair texels, the mask's jitter footprint at 0.5 scale, and RR/FSR history locking in residual dark samples; A/B RenderVKRTShadowMaskUpsampleTolMM 10 / 5).

Owner test of the fix (c499b1a318 / cc6af06e47, on main and release/1.5): clearly improved under all settings; native looks clean. Some shadow noise remains under DLSS Ray Reconstruction and slightly under FSR. Keeping open for the remaining upscaler-amplified noise (candidates: the temporal pass's plane test still accepting scalp history for hair texels, the mask's jitter footprint at 0.5 scale, and RR/FSR history locking in residual dark samples; A/B RenderVKRTShadowMaskUpsampleTolMM 10 / 5).

sky_line self-assigned this

2026-10-26 02:58:27 +00:00

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

sky_line/Slipstream#1086

No description provided.

Rows
Columns