Refocused 1.0.3 — film grain restoration, and RTX 50-series support on Windows
Refocused 1.0.3 is out now for all platforms. This release adds film grain restoration to the enhance and refine workflows, and fixes Windows builds for NVIDIA's Blackwell family of GPUs.
Film grain restoration
Upscalers tend to scrub away film grain, and with it, a lot of what makes footage feel like footage. 1.0.3 adds a Film Grain section to the enhance and refine settings for Refocus, Regenerate, and Regenerate-Q jobs.
Instead of synthesizing generic noise, Refocused isolates the actual grain from your source using an edge-preserving separation, scales it to the output resolution, and merges it back after all other processing. The result keeps the character of the original stock — not a simulated overlay.
- Grain restore strength runs from 0 (off) to 100% (the source's full grain), and can be overdriven up to 200% for a heavier look.
- Multi-stage pipelines do the right thing. When several passes are chained, the grain is restored from the original source footage at the final stage — not from an intermediate render that has already been cleaned.
RTX 50-series (Blackwell) support on Windows
The Windows GPU runtime has been upgraded across the board — TensorRT 10.9, CUDA 12.8, and a matching PyTorch runtime — with native kernels for NVIDIA's RTX 50-series GPUs. Building model engines locally and running every workflow now works out of the box on Blackwell cards, and prebuilt engine bundles have been rebuilt on the new runtime.
If you've previously built engines locally, the app will detect that they were built by the older runtime and offer to rebuild them — no manual cleanup needed. GPUs back to the GTX 16/RTX 20-series (Turing) remain supported, with no driver-requirement change in practice, although the Essentials Bundle prebuilt models still require Ampere+ GPUs.
Windows fixes
- Fixed an installation failure when another
taris on your PATH. Tools like Git for Windows ship their owntar.exe, which could break model and engine package extraction. Refocused now always uses the Windows built-in. - Fixed the app shipping an older CUDA runtime than the bundled GPU libraries expected.
To update, download the latest build from the install page — or use the in-app updater in Settings.