A new open-source project on GitHub, frameglide, aims to convert 30fps video streams from consoles such as the PS5 into 60–120fps output on Windows PCs using AI-based frame interpolation. Designed for remote-play setups, the tool enhances visual fluidity without requiring changes to the source console. It runs on the client side, leveraging real-time neural rendering with PyTorch and Vulkan, and supports integration with streaming clients like Chiaki.
How frameglide works under the hood
frameglide applies video frame interpolation—using AI to generate intermediate frames between existing ones—to increase perceived frame rate. It processes a decoded 30fps video stream, likely from a compressed network feed, and employs optical flow estimation with deep learning models such as RIFE (Real-Time Intermediate Flow Estimation) to synthesise new frames. This processing occurs entirely on the receiving PC, requiring a powerful GPU; NVIDIA RTX series cards are recommended due to their tensor cores optimised for AI inference.
The repository includes tags like DLSS, super-resolution, and upscaling, but frameglide is not affiliated with NVIDIA and does not implement DLSS 5 or any official DLSS technology. Unlike DLSS Frame Generation, which operates within the game engine and requires developer support, frameglide works externally on the video stream after decoding. This allows broad compatibility across games but introduces additional latency from frame buffering, AI inference, and re-rendering.
The tool uses Vulkan for GPU compute, enabling potential support for AMD GPUs, though performance without tensor cores may be limited. It does not modify the original stream protocol and is not a network-level optimisation.
Performance trade-offs and hardware demands
Real-time frame interpolation from 30fps to 60–120fps places high demands on GPU resources. While the project does not list minimum specifications, comparable tools typically require at least an RTX 3060 or equivalent for stable performance. Sufficient VRAM, memory bandwidth, and efficient PyTorch execution are essential. Users may experience increased power draw and thermal output, particularly when running alongside streaming applications.
Latency is a key constraint. Optical flow methods require at least two consecutive frames to estimate motion, introducing a buffer delay. In fast-paced gameplay, this added latency may diminish the benefit of smoother visuals. Interpolation quality also varies with scene complexity—rapid motion, transparency effects, or occlusions can lead to artefacts such as ghosting or blurring.
Network stability and source stream resolution affect input quality; poor initial streams reduce the effectiveness of AI enhancement.
Implications for remote gaming and cloud services
frameglide reflects growing interest in enhancing remote-play experiences beyond official capabilities. Sony’s PS Remote Play, for example, often delivers 30fps even on strong connections, limiting responsiveness. While companies like NVIDIA (DLSS 3+) and AMD (FSR 3) develop proprietary frame generation, frameglide offers an open, third-party alternative that could extend the life of existing hardware.
In the UK, UAE, US, and Kazakhstan—regions with rising adoption of remote gaming due to high broadband penetration and distributed lifestyles—such tools may attract technically skilled users looking to avoid console upgrades or cloud subscriptions. However, the cost shifts from recurring fees to upfront investment in capable local hardware.
What it means
For IT professionals and founders, frameglide illustrates the potential of edge-side AI to improve user experience on bandwidth-constrained streams. It underscores a broader trend: moving intelligence to the endpoint to compensate for infrastructure limitations. However, the project lacks a stated privacy policy or telemetry disclosure, raising questions about data handling—though processing occurs locally, users should verify that no data is exfiltrated.
Businesses developing streaming platforms might explore similar techniques to differentiate quality of service, but must weigh challenges around latency, licensing of AI models, and dependency on high-end client hardware.
Users should treat frameglide as experimental. It has no official support, security audit, or warranty. Competitive gamers should test latency impact thoroughly before adoption. The tool does not replace native high-frame-rate streaming but serves as a workaround for smoother visuals in non-latency-critical scenarios.