NVIDIA's DLSS Transformer model officially out of beta
, by Quentin
NVIDIA has announced that its Transformer-based DLSS model is no longer in the testing phase. It was shown at CES 2025 as part of the DLSS 4 update, which also includes improvements to ray reconstruction, super resolution and DLAA.
Unlike previous versions, this model treats the image as a whole and evaluates the importance of each pixel, even across multiple images. It uses twice as many variables as previous upscaling models.
According to NVIDIA, this results in greater pixel stability, less ghosting, improved detail in motion and higher quality in scenes with complex lighting. The transform model also improves ray reconstruction.
After six months in beta, this release marks a new milestone for future DLSS developments.
Medias
Source: github.com