RIFE video frame interpolation with NVENC encoding (AV1 / HEVC / H.264), wrapped in a simple desktop GUI.
FrameMeld reuses Flowframes' RIFE CUDA implementation for interpolation, then encodes the result straight to AV1/HEVC/H.264 using NVIDIA NVENC via FFmpeg — no manual frame extraction, no juggling separate tools.
Nvidia GPU required. Interpolation currently runs on RIFE CUDA, and encoding uses NVENC — both are Nvidia-only. AMD/Intel support (via RIFE-NCNN + AMF/QSV) is on the roadmap.
Features
RIFE CUDA interpolation (2x / 4x)
Encode to AV1 NVENC, HEVC NVENC, or H.264 NVENC
Configurable preset & CQ
Auto video info preview (resolution, framerate, frame count, duration) on input selection
Auto-generated output filename ({name} {fps}fps.{ext}) — just pick a folder and format
Clean, filtered log output (no raw command dumps or FFmpeg banner spam)
First-launch setup downloads the required runtime (FFmpeg + Python + RIFE model) automatically — no separate installs needed
Temporary working files are cleaned up automatically after a run finishes or is cancelled
Requirements
An NVIDIA GPU with NVENC support for your target codec (AV1 NVENC requires RTX 40-series or newer; HEVC/H.264 NVENC support goes back further — check NVIDIA's encoder support matrix).
Internet connection on first launch, to download the runtime (see How the runtime works below).
🚀 Download FrameMeld
Ready-to-use Windows build
Download FrameMeld from either of the links below:
If you prefer downloading the application through itch.io.
How the runtime works
FrameMeld itself is a small download. The heavy dependencies — FFmpeg, a Python distribution with PyTorch/CUDA, and the RIFE model — are fetched on first launch and stored in:
%LOCALAPPDATA%\FrameMeld\runtime\
This keeps the initial download small and lets FrameMeld update those components independently of the app itself. The app won't let you start a job until FFmpeg, the Python runtime, and an AI model are all downloaded.
These files are mirrored (as .7z archives) from their original sources on Hugging Face — see Credits & Licenses for where each component actually comes from.
Building from source
Only needed if you want to modify FrameMeld, verify the build, or produce your own release. If you just want to use the app, use the Download section above instead.
Requirements for building:
Python 3.11+ available as py on your system PATH.
That's the only build-time requirement — building FrameMeld just compiles the GUI itself with PyInstaller. It does not need FFmpeg downloaded locally; those are fetched by the app at runtime (see How the runtime works), not baked in at build time.
Steps:
Clone this repo.
Run Build_FrameMeld.bat.
The script will:
Create a virtual environment, install PySide6 + PyInstaller
Build dist/FrameMeld/FrameMeld.exe, bundling in app/tools/7za.exe (used later to extract the downloaded runtime archives)
The built app in dist/FrameMeld/ is ready to run as-is — launch FrameMeld.exe and it'll prompt for the runtime download on first run, same as a downloaded release. This dist/FrameMeld/ folder is also what gets zipped and published to Releases/itch.io.
Repository contents
This repo only tracks source code — no bundled runtime binaries or models:
app/FrameMeld.py source code
app/tools/7za.exe bundled archive tool (used to extract the downloaded runtime)
Build_FrameMeld.bat build script
README.md, LICENSE, THIRD_PARTY_LICENSES.md
dist/, build/, and .venv/ are generated locally when you build and are not committed — excluded via .gitignore. %LOCALAPPDATA%\FrameMeld\runtime (the downloaded FFmpeg/Python/RIFE files) lives outside the repo entirely, on the end user's machine.
Usage
Launch FrameMeld.exe. On first run, download FFmpeg, the Python runtime, and an AI model from the Setup section at the top.
Click Input and select a video. Its resolution, framerate, frame count, and duration will appear automatically.
Click Output Folder and pick where the result should be saved.
Choose an output Format (.mkv / .mp4 / .mov / .webm).
Set Multiplier (2x/4x), Scale, Encoder, Preset, and CQ as needed.
Click START. The output filename is generated automatically as {input name} {new fps}fps.{format}.
Roadmap
🧠 Interpolation Engines
RIFE CUDA
RIFE NCNN/Vulkan
DAIN
XVFI
Additional RIFE models
🎞️ Video Encoders
NVIDIA NVENC
AMD AMF
Intel QSV
Software encoding fallback
⚙️ Runtime & Distribution
Automatic runtime download
Per-user LocalAppData runtime
Hugging Face runtime mirrors
Runtime version management
Runtime integrity/hash verification
Automatic runtime updates
🖥️ Application
Video metadata detection
Automatic output naming
Temporary file cleanup
Batch processing
Progress estimation
Queue system
Credits & Licenses
FrameMeld is licensed under GPL-3.0 (see LICENSE), as it builds on GPL-3.0-licensed code from Flowframes.
