Flowframes

Flowframes

Create smoother, higher-frame-rate video with AI-generated frames. Flowframes turns 24 or 30 FPS footage into fluid 60, 120 or custom-frame-rate video using RIFE and other advanced interpolation models - all processed locally on your own GPU.

Download for WindowsHow it works

Free download · Flowframes 1.36.0 · Windows 10 & 11 · build options

Windows 10 & 11 Open-source donationware Nvidia, AMD & Intel GPUs

Smoother motion, generated between the original frames

Flowframes analyses the motion between existing frames and generates new ones that follow that movement - for a higher frame rate, smoother motion and cleaner slow motion. Unlike simple frame blending, AI interpolation reconstructs where objects should be at each new moment in time, preserving clarity that conventional resampling tends to lose.

Animated demonstration of Flowframes smoothing video with AI-generated intermediate frames
Before-and-after comparison of AI video frame interpolation in Flowframes

Built for real video workflows

Made for video editors, animators, visual artists, gamers and restoration workflows - anyone who needs better motion without rebuilding a project frame by frame.

Convert 24 or 30 FPS to 60 FPS

Increase the frame rate of films, edits, gameplay captures, screen recordings and generated video while keeping the original duration.

Create smoother slow motion

Generate extra frames before slowing footage down in your editor, for cleaner slow-motion that does not rely on duplicated or blended frames.

Interpolate 2D animation

Detect and remove repeated frames before interpolation to get smoother motion from anime, motion graphics and content drawn on twos or threes.

Complete 3D renders faster

Render alternate frames from a 3D scene, then generate the in-between frames with Flowframes to cut rendering time on suitable shots.

Improve AI-generated video

Smooth clips produced at low or inconsistent frame rates. Scene detection and de-duplication make Flowframes a natural fit for generative-video workflows.

Create seamless animated loops

Loop interpolation generates frames between the final frame and the first frame so an animated loop returns smoothly to its starting point.

Process image sequences

Import PNG, JPEG and other supported image sequences directly, then export the result as a video, animated image or new frame sequence.

How Flowframes works

Three steps from source file to finished, higher-frame-rate video. No separate editor required.

1

Add a video or image sequence

Drop in a supported video, GIF or sequence of images. Flowframes reads common formats including MP4, MKV, MOV, WEBM, GIF, PNG and JPEG.

2

Choose the interpolation settings

Pick an interpolation factor or target frame rate, choose the AI model, and adjust optional controls. The defaults work well for many videos.

3

Export the result

Encode to MP4, MKV, WEBM, MOV, GIF or an image sequence. Audio, subtitles and compatible metadata can be carried over from the source.

RIFE, DAIN and more - built around the best models

Flowframes is designed around RIFE, a fast AI model for frame interpolation, with additional and experimental models available depending on the build. For most people, RIFE is the recommended start.

RIFE CUDA

Recommended for compatible Nvidia GPUs when speed is the priority. CUDA processing can use half-precision acceleration to reduce VRAM use and run faster on supported hardware.

RIFE NCNN (Vulkan)

Runs through the NCNN framework using Vulkan, for broad compatibility across supported Nvidia, AMD and Intel graphics - no Nvidia CUDA environment required.

DAIN NCNN

An alternative interpolation model available through Vulkan-compatible NCNN processing. Useful for testing different motion characteristics and comparing against RIFE.

FLAVR & XVFI

Experimental model options in supported builds. They provide additional approaches for specific footage, testing and advanced interpolation workflows.

Made for Nvidia, AMD and Intel graphics

Flowframes does not require an Nvidia GPU for standard NCNN processing - Vulkan-based models run on supported graphics hardware from multiple vendors.

Nvidia GPU

Use RIFE CUDA for the fastest processing on supported Nvidia hardware, or run the Vulkan-based NCNN implementation.

AMD GPU

Use RIFE NCNN or another available NCNN model through Vulkan. An Nvidia GPU is not required to use Flowframes.

Intel GPU

Modern Vulkan-capable Intel graphics can use the NCNN path. Performance and compatibility depend on the specific GPU and driver.

Actual speed depends on the model, source resolution, interpolation factor, GPU performance, available VRAM and encoding settings.

Smart processing for clean results

Frame interpolation is only as good as how it handles duplicates, cuts and audio. Flowframes covers all three so you spend less time fixing artifacts.

Intelligent frame de-duplication

Many 2D animations hold the same drawing across several frames. Flowframes can detect and remove those duplicates before interpolation, so the model works with real movement instead of identical images - and it stays off for camera footage and gameplay.

Scene-change protection

A hard cut has no continuous motion across it. Flowframes detects scene changes and avoids generating frames across recognised cuts, preventing the morphing artifacts that ruin automated interpolation. Sensitivity is adjustable.

Keep audio, subtitles & metadata

Raising a video’s frame rate should not mean rebuilding its audio. Flowframes can transfer audio tracks, subtitle tracks and compatible metadata from the source, copying streams without re-encoding where the container allows.

Flexible input and output

Import common production and distribution formats, then export to the container and codec you need.

Input

MP4MKVMOVWEBMGIFBIKPNG sequencesJPEG sequences

Output

MP4MKVWEBMMOVGIFPNG / JPEG sequences

Export codecs include H.264, H.265 / HEVC, VP9, AV1, subject to the selected encoder and hardware configuration.

Advanced control, without the busywork

Sensible defaults let you start immediately. These controls are there when a project needs them.

Automatic encoding

Encode the output while interpolation is still running, and optionally delete processed frames to save temporary storage.

High-resolution mode

RIFE UHD Mode adjusts model scaling for high-resolution footage; a maximum processing resolution can also be set for speed.

Custom output frame rates

Generate a multiplied frame rate or cap the export to a specific maximum, such as 24 FPS in to exactly 60 FPS out.

Transparent interpolation

Preserve and interpolate alpha transparency when the input, intermediate and output formats all support it.

Loop interpolation

Generate motion from the end of an animation back to its first frame to create a smoother repeating loop.

JPEG intermediates

Use JPEG intermediate frames to reduce storage use and speed up extraction when lossless PNG is unnecessary.

Manual processing stages

Run extraction, interpolation and encoding as separate steps when frames need to be inspected, edited or removed.

Multi-GPU configuration

Assign one or more supported GPUs by device ID for compatible processing modes and hardware setups.

Custom FFmpeg encoding

Configure codec, quality, bitrate, container and other export parameters with advanced FFmpeg options.

Custom file naming

Define output-name patterns from project and source variables for more consistent file organisation.

Download Flowframes for Windows

Free, open-source AI frame interpolation that runs on your own GPU - no uploads, no per-video fees, no cloud queues. Built for Windows 10 and 11.

Installer (.exe) · Flowframes 1.36.0 · Windows 10 & 11

Flowframes is open-source donationware: the newest early-access builds reach supporters first, and the source is always public. Want the very latest version? Support development and get early-access builds on Patreon.

System requirements

Flowframes runs interpolation locally on the Windows PC where it is installed. Older graphics hardware may lack the Vulkan features NCNN needs.

Minimum

  • Windows 10 or Windows 11
  • A Vulkan-capable graphics card
  • A supported Nvidia, AMD or Intel graphics driver
  • Storage for the source, temporary frames and exported result

Recommended

  • A modern CUDA-capable Nvidia GPU
  • 6 GB of VRAM or more
  • 16 GB of system memory
  • A modern multi-core Intel Core or AMD Ryzen CPU
  • Solid-state storage for frame extraction and encoding

Not supported: GPUs such as the Nvidia GT 710 and Intel HD 3000/4000 series, and Windows 7.

When frame interpolation works best

AI interpolation performs best when movement is visible and reasonably predictable between adjacent frames. It creates plausible new frames - it cannot recover detail that was never in the source.

Strong use cases

  • Camera movement with consistent motion
  • Gameplay and screen captures
  • Character animation
  • Anime and hand-drawn animation (after correct de-duplication)
  • CGI and 3D rendering
  • Motion graphics
  • Slow-motion preparation
  • AI-generated clips with low frame rates

Harder for any model

  • Heavy motion blur
  • Large objects appearing or disappearing between frames
  • Fast cuts that are not detected correctly
  • Flashing lights or rapid exposure changes
  • Reflections, particles, smoke, water or transparent objects
  • Severe compression artifacts or corrupted source timing

Frequently asked questions

What is Flowframes?

Flowframes is a Windows application for AI-powered video frame interpolation. It analyses the motion between existing frames and generates new intermediate frames to create smoother playback, higher frame rates and more natural slow motion. It is built around advanced models such as RIFE and supports both CUDA and Vulkan-based processing.

What can Flowframes be used for?

Flowframes can increase the frame rate of video, prepare footage for slow motion, smooth 2D animation, improve AI-generated clips, interpolate gameplay recordings, create seamless animated loops and generate missing frames for suitable 3D rendering workflows. It works with complete videos, animated GIFs and image sequences.

Does Flowframes increase video resolution?

No. Flowframes increases temporal resolution by generating extra frames between the original frames. It does not increase the width, height or spatial detail of the source. A 1080p video stays 1080p - the difference is that motion looks smoother because more frames are shown each second.

Can Flowframes convert 24 or 30 FPS video to 60 FPS?

Yes. Flowframes can generate additional frames and export at a higher frame rate, including common targets such as 48, 60, 120 or 240 FPS. You can also set a maximum output frame rate - for example, generating frames from a 24 FPS source and downsampling the result to exactly 60 FPS.

Which AI model should I use?

RIFE is the recommended starting point for most videos, with a strong balance of quality, speed and hardware compatibility. Nvidia systems can use RIFE CUDA for faster processing; RIFE NCNN uses Vulkan and supports a wider range of GPUs. DAIN, FLAVR and XVFI are available for comparison and specific footage.

What is the difference between RIFE CUDA and RIFE NCNN?

RIFE CUDA is designed for compatible Nvidia graphics cards and is generally the faster option on supported hardware. RIFE NCNN runs through Vulkan and offers broader GPU compatibility, making it the right choice for AMD systems and for Nvidia systems that do not use the CUDA path. The intended result is similar; performance, memory use and model availability may differ.

Does Flowframes support AMD GPUs?

Yes. Compatible AMD GPUs can use Vulkan-based NCNN models such as RIFE NCNN. CUDA modes are exclusive to Nvidia hardware, but an Nvidia GPU is not required to use Flowframes - the graphics card just needs to support the Vulkan features the selected NCNN model requires.

Does Flowframes require additional software?

Standard builds include the components required for normal use, and the installer can download necessary dependencies automatically. Most people do not need to install Python, PyTorch, FFmpeg or AI models by hand. Advanced users can configure a system Python environment or compile the open-source project themselves.

Which video and image formats are supported?

Flowframes reads widely used formats such as MP4, MKV, MOV, WEBM, GIF and BIK, and can import image sequences in PNG and JPEG. Results export as MP4, MKV, MOV, WEBM, GIF or a sequence of frames, with codecs including H.264, H.265/HEVC, VP9 and AV1 depending on the encoding configuration.

Will Flowframes preserve audio and subtitles?

Flowframes can transfer supported audio tracks, subtitle tracks and compatible metadata from the source to the exported video. Streams can often be copied without quality loss when the chosen output container supports them; compatibility depends on the source format, output format and encoding settings.

What is frame de-duplication?

Frame de-duplication detects and removes consecutive duplicate frames before interpolation. It is mainly for 2D animation, anime and motion graphics that intentionally hold the same image across several frames. Removing those repeats lets the model work with real movement instead of identical images. It should usually be disabled for camera footage, gameplay and 3D-rendered video.

Why does an interpolated video sometimes look choppy?

Choppy results are often caused by incorrect frame de-duplication. Dark, low-contrast or slowly changing frames can be wrongly detected as duplicates and removed. For camera footage, gameplay and most CGI, de-duplication should be off; for animation, lower the detection threshold or use the more accurate post-extraction mode. Irregular source timing and compression damage can also affect the result.

How does Flowframes handle scene changes?

Flowframes includes scene-change detection so it does not generate intermediate frames across hard cuts. Without it, the model may try to blend two unrelated shots and produce a visible morphing artifact. Detection keeps the end of one shot separate from the start of the next, and sensitivity can be adjusted for fast editing, fades or flashes.

Is Flowframes free and open source?

Yes. Flowframes is open-source donationware. The complete source code is publicly available for inspection, contribution and independent compilation. New compiled builds reach supporters first before selected versions become free public releases, but the source itself is not locked behind a payment.

Is the Flowframes download here safe?

The download on this page is the genuine Flowframes installer for Windows. This is an independent resource site that makes the app easy to find; we do not repackage, wrap or modify the software. For the official project, source code and the newest early-access builds, see the About page.

What are the system requirements for Flowframes?

Flowframes needs Windows 10 or 11 and a Vulkan-capable graphics card. A modern CUDA-capable Nvidia GPU with at least 6 GB of VRAM is recommended for faster processing, along with 16 GB of RAM and solid-state storage for high-resolution or long projects. Very old hardware such as the Nvidia GT 710 and Intel HD 3000/4000 series is not supported, and Windows 7 is not officially supported.

Make every frame count

Generate smoother motion, prepare cleaner slow motion and interpolate animation - from one focused Windows app. Free, open source, and processed entirely on your own GPU.

Download Flowframes for WindowsHow it works