AO (SSAO and GTAO): visible lines on the ground in some areas #648

New issue

Closed

opened 2026-10-10 22:04:09 +00:00 by sky_line · 2 comments

sky_line commented

2026-10-10 22:04:09 +00:00

Owner

Visible lines show up on the ground in some areas. They are clearest on the grass terrain area that also showed banding earlier. Both AO methods are affected: SSAO (RenderVKAOMethod 0) and GTAO (1). Turning ambient occlusion off (RenderDeferredSSAO 0) should remove them. That needs confirming as part of the fix.

Because both kernels show it, the cause is probably in what they share, not in either kernel. Suspects:

Depth precision or reconstruction on large, gently sloped terrain. The lines would then be false self-occlusion along depth-quantization steps.
Normals the AO reads: G-buffer encoding, or flat-interpolated terrain normals that create facet lines.
The shared blur/upsample: depth-aware weights breaking along terrain triangle edges.
Missing bias or thickness in the horizon test at grazing angles.

Deferred; we'll fix it later. Repro: stand on the grass area where banding was seen before, with AO on and either method selected.

Visible lines show up on the ground in some areas. They are clearest on the grass terrain area that also showed banding earlier. **Both AO methods are affected**: SSAO (`RenderVKAOMethod` 0) and GTAO (1). Turning ambient occlusion off (`RenderDeferredSSAO` 0) should remove them. That needs confirming as part of the fix. Because both kernels show it, the cause is probably in what they share, not in either kernel. Suspects: - Depth precision or reconstruction on large, gently sloped terrain. The lines would then be false self-occlusion along depth-quantization steps. - Normals the AO reads: G-buffer encoding, or flat-interpolated terrain normals that create facet lines. - The shared blur/upsample: depth-aware weights breaking along terrain triangle edges. - Missing bias or thickness in the horizon test at grazing angles. Deferred; we'll fix it later. Repro: stand on the grass area where banding was seen before, with AO on and either method selected.

sky_line added the

rendering

bug

labels

2026-10-10 22:04:09 +00:00

sky_line added this to the Slipstream project

2026-10-10 22:04:09 +00:00

sky_line commented

2026-10-11 03:19:45 +00:00

Author

Owner

New finding. AO strength changes with camera distance: a pony face darkens noticeably when zoomed out and loses that darkening up close. The cause is the screen-space radius cap. GTAO uses RenderVKGTAORadius 0.5 m capped at RenderVKGTAOMaxPixels 192 px; SSAO uses RenderSSAOScale / RenderSSAOMaxScale 200 px. Close to the camera the cap shrinks the world radius in proportion to distance, so the large-scale occlusion vanishes. This may also be the cause of the ground lines, since the effective radius varies across a large terrain plane. An agent is working on a distance-invariant radius using depth MIPs instead of the cap. If that fixes the lines too, this ticket closes with it.

New finding. AO strength changes with camera distance: a pony face darkens noticeably when zoomed out and loses that darkening up close. The cause is the screen-space radius cap. GTAO uses `RenderVKGTAORadius` 0.5 m capped at `RenderVKGTAOMaxPixels` 192 px; SSAO uses `RenderSSAOScale` / `RenderSSAOMaxScale` 200 px. Close to the camera the cap shrinks the world radius in proportion to distance, so the large-scale occlusion vanishes. This may also be the cause of the ground lines, since the effective radius varies across a large terrain plane. An agent is working on a distance-invariant radius using depth MIPs instead of the cap. If that fixes the lines too, this ticket closes with it.

sky_line added the

status

in-progress

label

2026-10-11 03:19:45 +00:00

sky_line referenced this issue

2026-10-11 04:00:16 +00:00

RT shadows: shadow terminator artifacts on smooth-shaded low-poly mesh (triangles fill stepwise) -- Hanika offset for all RT shadow rays #664