02 / A SECOND WORLD FIRST*
DLSS5 Emulation
ForgePlay applies neural visual enhancement to completed RGB frames of Windows games running on Mac. It uses publicly available HyPER-GAN weights to approximate a DLSS 5-like appearance—not NVIDIA DLSS code or weights.
The contribution is the integration into the Windows-game execution and presentation path on Mac—not the invention of HyPER-GAN or a port of NVIDIA DLSS 5.
In 2.0.0, DLSS5 Emulation and frame generation can run together and be switched independently. Both default to off in new configurations; saved choices are preserved. Performance varies by game, resolution, and settings.
02 / DLSS5 Emulation — scope, timeline, and sources
What “world first” means here
The developer’s claim concerns real-time integration of HyPER-GAN-based photorealistic post-processing into completed frames of Windows games running on macOS, without NVIDIA DLSS code or weights. The public-source review dated 27 September 2026 found no earlier implementation within that scope. It does not certify every unpublished or unindexed implementation worldwide.
19 September is the MVP results publication date. 27 September is the prior-work review date. Neither is an inferred date for the original idea or the precise end of MVP testing.
From development to public release
Before the results publication · exact date not established
Concept and development
The developer designed, implemented, and tested the MVP before publishing its results. No exact concept or test-completion timestamp is assigned without a separate record.
MVP tested · results published
The developer confirms that MVP testing was already complete when the notice was published. The update announcement and before-and-after images document the public disclosure of a working MVP, not only a development plan.
- Notice commit
- Results-image commit
These are publication-change timestamps, not the start or end of testing. The developer’s confirmation establishes the MVP’s status; the commits and successful deployments establish its publication history.
After the MVP · before the release
Refinement and product integration
The published MVP was refined into 2.0.0: color processing, configurable processing passes, and integration with frame generation. This does not move the completed MVP back to a planning stage.
Public release · 2.0.0 (Build 6)
Released on GitHub at 03:21:26 KST on 27 September 2026 (18:21:26 UTC on 26 September). The release date is distinct from the MVP results publication.
Release record and downloads ↗
Related work and the scope of comparison
Photorealistic enhancement and HyPER-GAN integration already exist in other contexts. The claim is narrower: their integration into Windows games running on macOS. The following original sources should be read within their own scope.
- HyPER-GAN ↗
- The original model and pretrained weights, with CARLA real-time inference and Unreal Engine integration instructions. ForgePlay does not claim to have invented the model or its first real-time use.
- CARLA2Real ↗ / Hybrid-Sim2Real ↗
- Photorealism enhancement for CARLA simulation and game-engine synthetic datasets. This is a different application scope from integrating post-processing into Windows games on Mac.
- EndoLive ↗
- An endoscopic style-transfer project using ConStructS and HyPER-GAN that reports operation on Apple Vision Pro. It is not a Windows-game integration, and shows why “first HyPER-GAN use on Apple hardware” would be too broad a claim.
The supplied fork comparisons were also checked. On 27 September 2026, lastquest/main was identical to upstream/main; porom004/main was five commits behind with no ahead commits. These comparisons alone neither prove nor disprove worldwide priority.
The 19 September notice, captions, and images remain unchanged historical records. They do not independently establish FPS, latency, NPU placement, or a model checkpoint. Current specifications belong to the 2.0.0 release notes.