Hey guys, after spending quite some time experimenting with temporal reconstruction in ReShade, I'd like to share this shader,The original goal wasn't to build "another TAA shader", but to see how far a purely image-space temporal filter could go within ReShade's limitations.Since ReShade doesn't have easy access to motion vectors or engine depth history, TDAA relies entirely on image analysis and adaptive history validation instead of conventional reprojection.Current features include:
- Adaptive temporal history blending
- YCoCg variance clipping
- Neighborhood history search (up to 7×7)
- Disocclusion detection
- Selective Laplacian detail recovery
- Noise-aware sharpening
- Debug visualization
The shader is designed to work
together with a spatial AA solution such as
CMAA2 or
SMAA, rather than replacing it.The overall philosophy is different from most TAA implementations:instead of maximizing temporal accumulation (which usually leads to progressively softer images), TDAA tries to preserve fine detail such as text, foliage, fabric patterns and thin geometry while still reducing temporal instability.It's obviously still constrained by what ReShade can access, so there are situations where engine-native TAA with motion vectors will always have an advantage. However, I think the current results are surprisingly competitive considering those limitations.Performance mainly depends on the selected search radius, with 5×5 generally offering the best balance between quality and GPU cost.
(Initially i wanted to reduce shimmering but i ended up making a broken pseudo smaa before approaching this method) GitHub repository:
github.com/aston89/TDAA-Tempodal-Detail-...Aliasing-for-ReShade