Offline AI frame interpolation for NVIDIA GPUs.
Drop in a video, pick a target frame rate, and get a buttery-smooth high-FPS copy,
no cloud, no subscription, no install.
The same shot, before and after. Left: a classic anime-style pan at 10 fps, the cadence anime pans are actually drawn at. Right: the same clip interpolated 5× to 50 fps. Every original frame passes through untouched; the AI draws only the frames in between.
It also brings 108-year-old anime back to life. Left: Namakura Gatana (1917, public domain), 15 fps with a century of film grain. Right: the same frames after AI detail restoration, 2× RTX upscale and 15→50 fps interpolation, one render.
Source: the National Film Center scan, via the Internet Archive (Public Domain Mark).
Built because NVIDIA's RTX Video (AI upscaling + TrueHDR) is gorgeous but playback-only: it enhances what you watch, then throws it away. Smooth My Video applies those same RTX passes (plus AI frame interpolation) straight to your file, so the result is saved, not just streamed.
Pick a video (or drag it in), choose a target frame rate (double it, 4×, 8× and up, or match your monitor's refresh), and click Smooth It!. Smooth My Video generates the in-between frames with a GMFSS AI model on your GPU and writes a smoother, high-frame-rate copy right next to the original. In the same render it can also upscale (up to 16K), sharpen, restore detail, and convert SDR → real HDR10, while carrying over every audio track, subtitle, chapter and font.
Built and tested on an RTX 5090 Laptop; runs on any recent NVIDIA GPU with a current driver.
- 🎞️ Smooth and sharp. Your real frames pass through at full quality with AI-generated frames woven in between, you get the higher frame rate without softening or reprocessing the original footage.
- 🧊 10-bit output by default. Float-precision interpolated frames are written at 10-bit, so smooth gradients (skies, glows) never band into visible steps.
- 🎨 Production-grade HDR10. Real SDR→HDR10 conversion with proper mastering metadata (mastering-display + measured MaxCLL/MaxFALL) and faithful, cyan-free colour, not just a PQ tag.
- 🌈 Dolby Vision Profile 8.1 export (experimental). Optionally add Profile 8.1 dynamic-HDR metadata on top of the HDR10 render, HDR10-compatible, so non-DV players fall back to HDR10. Uses the separately-installed open-source dovi_tool; no Dolby software is bundled.
- ➕ HDR10+ export (experimental). Optionally embed HDR10+ (SMPTE ST 2094-40) dynamic-HDR metadata, measured per frame from the actual render, also HDR10-compatible, and combinable with Dolby Vision. Uses the separately-installed open-source hdr10plus_tool.
- 🔍 AI upscaling to 16K + detail restoration. NVIDIA RTX Video Super Resolution plus a Real-ESRGAN restore pass, layered with the interpolation in a single render.
- 💬 Keeps every track. All audio, subtitles/translations, chapters and font attachments are preserved
(auto-switches to
.mkvwhen needed), nothing silently dropped. - 🗜️ Visually lossless. HEVC / AV1 / H.266 encodes at maximum encoder effort, tuned and verified against a lossless 8K master (VMAF ~99.8, SSIM ≥ 0.995), no fiddly quality knob to guess at.
- 📦 100% offline & self-contained. Extract the zip and run, no Python, no pip, no ffmpeg to install, no account, no cloud upload. Only the NVIDIA driver is assumed. Free.
- ⚡ Fast. fp16 with a TensorRT backend, built and cached per resolution; 4K sources interpolate at nearly 1080p cost (motion is estimated at a resolution-appropriate scale, automatically). The app also tells you when a laptop "Silent" power profile is throttling the GPU.
- 🧺 Set-and-forget batches. Queue many files; a file that fails is noted and the rest keep rendering, and a batch interrupted by a crash or restart is re-queued on the next launch.
- 📱 Handles real-world files. Variable-frame-rate sources (phone clips, screen recordings) are detected and timed correctly, so audio never drifts out of sync.
- 🔁 Reproducible. The same file with the same settings renders byte-for-byte identically, every time.
- ✨ Optional DLSS 4.5 model. One checkbox switches the interpolation to NVIDIA's DLSS Frame Generation, the AI frame generation from their game stack, hosted offline by a bundled bare-bones D3D12 presentation loop (DLSS-FG has no video API, so the app runs one for it). Fully bundled (~10 MB, NVIDIA-redistributable Streamline runtime); whole multipliers 2×–6× on the source grid (multi-frame generation; above 2× needs an RTX 50); needs an RTX 40/50 GPU, a recent driver and Windows hardware-accelerated GPU scheduling. One caveat: NVIDIA's RTX Video enhancement (Super Resolution / Video HDR) processing a video playing in a browser preempts frame generation. The render detects the stall within seconds and heals itself once the video is closed or paused; if the interference persists it stops cleanly with progress saved, and Resume continues from there (other models are unaffected).
- 🎬 A RIFE model for live action. GMFSS is an anime specialist; one checkbox switches to Practical-RIFE 4.26 heavy (bundled, nothing to install), the strongest open general-purpose interpolation model - the pick for filmed content. TensorRT-accelerated like GMFSS. Its Preserve anime pacing (DRBA) sub-option keeps characters at their original animation cadence while camera pans smooth fully, for purists who want fluidity without "hollywoodizing" the animation.
- 🌀 Optional NVIDIA Smooth Motion model. One checkbox switches the interpolation from the GMFSS AI model to NVIDIA's hardware optical-flow FRUC (the same family as the driver-level Smooth Motion), for when you want the NVIDIA look or a non-AI reference. Lower quality than GMFSS on fast motion; the NVIDIA runtime is a one-time separate download (not bundled).
- 🌊 Optional SVP model. One more checkbox renders with the SmoothVideo Project's svpflow engine (the tech behind SVP's real-time playback smoothing), borrowed from your local SVP 4 installation, so SVP fans can bake their player's look into a file. Its NVIDIA Optical Flow sub-option (on by default) takes motion vectors from NVIDIA's hardware optical-flow unit; turning it off uses SVP's classic block-matching search instead. Requires SVP 4 installed (only its svpflow plugin DLLs are used; SVP's optional mpv player component is not needed).
- 🎚️ Plus the essentials: FSR-style sharpening, a live before/after preview, every setting remembered between runs, and a quiet one-line notice when a newer release is out (nothing auto-downloads).
- Download & run: grab the latest
SmoothMyVideo-<version>-win.zip, extract it anywhere, and runSmoothMyVideo.exe(or the Desktop / Start-menu shortcut). No install, no dependencies, just a current NVIDIA driver. A sample clip ships insamples/test.mp4. Release notes and checksums live on the GitHub releases page. - From source:
npm start. See DEVELOPMENT.md for the full setup.
The UI is Electron + TypeScript; the interpolation runs in a Python GMFSS_Fortuna engine spawned as a subprocess. GMFSS_Fortuna is a "union" interpolator, GMFlow optical flow, an IFNet/RIFE refiner, plus MetricNet, FeatureNet, FusionNet and softsplat warping, producing clean frames even at high multipliers. It runs fp16 with a cupy softsplat kernel and an optional TensorRT backend.
📖 Build it, hack on it, or read the design rationale → DEVELOPMENT.md
Smooth My Video was built end to end by AI coding agents (Claude Code), and it's meant to keep growing that way. If there's a feature you want, open an issue and describe it, or send a pull request. Contributions are welcome whether you write the code yourself or hand the idea to an agent, exactly how the rest of this app was built.
Dolby and Dolby Vision are trademarks of Dolby Laboratories. Smooth My Video is an independent project and is not affiliated with, endorsed by, sponsored by, or certified by Dolby Laboratories. Dolby Vision Profile 8.1 metadata is produced by the separately-installed, third-party open-source dovi_tool; no Dolby software is bundled or redistributed. HDR10+ is a trademark of HDR10+ Technologies, LLC; Smooth My Video is likewise not affiliated with, endorsed by, or certified by HDR10+ Technologies. HDR10+ metadata is injected by the separately-installed, third-party open-source hdr10plus_tool; no HDR10+ LLC software is bundled or redistributed.