hzwer/Practical-RIFE

More practical frame interpolation approach.

What it solves

Practical-RIFE provides a practical implementation of video frame interpolation, allowing users to increase the frame rate of a video by generating intermediate frames. It focuses on subjective visual quality and practical usability for engineers and developers rather than just optimizing for mathematical indices like PSNR.

How it works

The project is based on the RIFE (Real-Time Intermediate Flow Estimation) and SAFA (Scale-Adaptive Feature Aggregation) architectures. It uses trained models to estimate the optical flow between existing frames and synthesize new frames to create smoother motion. It offers various pre-trained models, including "lite" versions for lower computational costs and specialized versions for anime scenes or diffusion-model-generated videos.

Who it’s for

Engineers and developers who need to perform high-quality video frame interpolation and video enhancement, specifically those who can work with Python and command-line interfaces.

Highlights

  • Multiple Pre-trained Models: Provides a wide range of versions (e.g., v4.25, v4.26) with options for lite versions and specialized optimizations for anime and AI-generated videos.
  • Flexible Interpolation: Supports 2X, 4X, or custom multipliers for frame rate increases.
  • High-Resolution Support: Includes a --UHD or --scale parameter to handle 4K videos efficiently.
  • Video Enhancement: Incorporates SAFA for space-time video super-resolution and enhancement.
  • Command-Line Interface: Simple execution via inference_video.py for processing videos or image sequences.

Related

  • Project
    cclank/lanshu-create-ai-presenter-video

    A generalized AI digital human video production workflow that transforms scripts and images into synchronized presenter videos with subtitles and effects.

  • Project
    huangserva/ComfyUI_MiniMaxH3_Director

    A collection of ComfyUI workflows for MiniMax H3 Director, enabling text-to-video, image-to-video, and advanced video-to-video editing with reference materials.

  • Project
    shinkuan/Akagi

    Akagi is a Rust‑based desktop assistant for online riichi mahjong (Mahjong Soul, Tenhou, etc.). It captures game traffic via a MITM proxy or Chromium DevTools, runs a built‑in neural‑net bot (and optional cloud‑inference model) to compute shanten, waits, win probabilities, opponent risk, and suggested discards, and shows these numbers in a draggable HUD. Completed matches are stored locally and visualised in a History tab with rank pie charts, PT trend lines, and detailed stats. The app is distributed as a single portable binary for Windows, macOS (Apple Silicon) and Linux, uses Tauri 2 + React for the UI, and is licensed under Apache 2.0.

  • Project
    jianjieyiban/JJYB_AI_VideoAutoCut

    A local-first AI video creation workstation that automates the workflow from material analysis and scriptwriting to voiceover generation and audio-visual synchronization.