3Minute Funscript Web Studio PRO Tour Video Link
Funscript Web Studio PRO v8.2
Turn a video into an editable .funscript on your own Windows PC.
The three-minute tour above follows one real clip through the whole app, in the English interface.
Updated September 2026 for v8.2. The original v6.0 post and its revision notes are folded at the bottom.
*SAM3 is most recommend mode but slow and need to GPU with SAM3 Model (Pack included)
Get it
Download: Patreon Link
Which file do I need?
- No NVIDIA graphics card, or not sure? Get Standard (
FunscriptWebStudioPRO_v8.2_Standard.zip, about 390 MB). It runs on any Windows 10/11 64-bit PC, and point tracking uses the CPU. - NVIDIA RTX 20 / GTX 16 series or newer? Get both GPU files:
..._GPU_App.zip(about 3.0 GB) and..._GPU_ModelPack.zip(about 2.1 GB). Extract them into the same folder. This adds TAPNext++, the steadiest point tracking. - Want SAM3 (High-performance)? It is not a separate download. Run
SETUP_SAM3.batin either edition. It needs an NVIDIA card with 8 GB or more video memory.
Quick tips for choosing a download
- To check your card, open Task Manager → Performance → GPU.
- The ModelPack does nothing on its own, and adding it to Standard does not enable TAPNext++.
- Only downloaded the App file? Run
DOWNLOAD_TAPNEXTPP.batin its folder to fetch the model. - Still unsure? Start with Standard. Your videos and saved scripts work in either edition.
What it does
Video → Target → Track → Review → Export. Each step shows only what it needs, so you can take a clip from the first frame to a saved script without hunting through menus.
Three ways to choose what moves
Points. Click up to 9 spots on the frame, with no box to draw. TAPNext++ runs on an NVIDIA GPU, and TAPIR is the CPU engine in the Standard edition.
ROI box. Choose Precision, Fast or AI Auto. Draw the box yourself, or let Suggest from motion in box tighten a rough one. Check ROI quality tests the box on 4 seconds of video before you commit. Long clips can be split into segments, each with its own box, and tracked in one run.
SAM3 (High-performance). No points and no box. A SAM 3-based workflow finds the person, follows the pelvis in 3D, and can align strokes to the beat. Amplitude expansion starts at 0%, so the tracked depth is kept unless you ask for more.
Review before you export
- Quality Center scores the run and lists regions worth a second look. The strip under the graph marks high, medium and review confidence. Scores guide review; they are not accuracy percentages.
- Box vs content confirms a Precision box stayed on the moving target.
- Manual editing with undo and redo, plus regional re-tracking for a bad stretch in the point and ROI workflows.
- Sync offset set by hand, or estimated from the audio.
Shape the result
- Presets (Gentle, Normal, Hard) with Preserve stroke depth, so gentle passages stay gentle and intense ones stay deep.
- Output curves: Exact v6 output, fewest points for easy editing, or smooth, impact-focused and dense curves.
- Auto Optimize, the Humanizer with Learn My Style, and fine tuning for softness, stroke range and vertical shift.
- Device optimization and speed limits for SR1/OSR, Handy and basic devices, plus Stroke Booster.
- Region filters: Shift-drag to select a stretch, then apply Boost, Denoise, Expand, Bezier smoothing and more. Every change can be undone.
Export and video library
Save a standard .funscript next to the video, use Open in OFS to keep editing in OpenFunscripter, or launch MultiFunPlayer for playback. The built-in Video Library has thumbnails, ratings, search and script badges. With FFmpeg installed you also get trimming, rotation and automatic audio sync.
Hardware requirements (measured on the shipping build)
| Standard | GPU | |
|---|---|---|
| Point tracking | TAPIR (CPU) | TAPNext++ (NVIDIA) |
| Graphics card | not required | RTX 20 / GTX 16 series or newer |
| Driver | — | CUDA 12.8 capable, 570.65 or newer |
| Video memory | — | 4 GB or more (measured 2.0 GB) |
| Disk | about 0.7 GB | about 7.5 GB |
Both editions run on Windows 10 or 11 (64-bit) and need no separate Python. The GPU edition still runs without an NVIDIA card: TAPNext++ switches itself off and points fall back to TAPIR.
SAM3 mode is a one-time setup with SETUP_SAM3.bat. It needs an NVIDIA GPU with 8 GB or more video memory (measured 6.8 GB, 12 GB comfortable), Python 3.11 from python.org with the py launcher, about 11 GB of disk (roughly 10 GB of downloads), and approved access to facebook/sam3 and facebook/sam-3d-body-dinov3. The first run on a short clip takes a few minutes; changing output options afterwards takes seconds.
Honest limits
A generated track still needs a visual check. Camera movement, occlusion and several people in frame can throw any tracker off, and SAM3’s depth comes from a single camera. The app flags what it can; the final call is yours.
Questions and feedback are welcome in this thread. English and Korean manuals are included.
Older versions (outdated)
v6.0 post (February 2026)
Funscript Web Studio v6.0
※ All of the woman photo is AI created by me.
A web-based funscript creation tool optimized for SR1 2.0 users
Web-Based · No VR/Multi-Axis · SR1 2.0 Optimized
Funscript Web Studio is a browser-based tool designed to make funscript creation faster, more flexible, and more precise. While it does not support VR or multi-axis devices, it has been carefully built over roughly two weeks to provide the best possible experience for SR1 2.0 users — covering tracking quality, feature depth, and workflow convenience.
8 Tracking Methods
The studio offers 8 distinct tracking algorithms. Rather than relying on a single AI-generated approach, each method is tuned for different scene types and content styles — so you can pick the one that best fits the situation.
9 Tracking Style Presets
Nine presets let you shape the output feel immediately, from slow and gentle to aggressive and high-speed.
Per-Segment Tracking with ROI
Long videos can be divided into segments, each with its own tracking box (ROI). This means different parts of a video can be tracked with completely different settings — ideal for content with changing scenes or camera angles. Scenes can even be detected automatically.
Fine-Tuning Sliders & 15 Post-Filters
After tracking, apply any of 15 per-segment filters to polish the result — from noise removal and curve smoothing to rhythm normalization and point reduction.
+Vibration + pattern nomalizing etc
Manual Editing & Re-tracking
Select any region of the script, choose a different tracking method, and re-track just that section — without touching the rest. This makes targeted fixes fast and non-destructive.
Save Funscript & Open in OFS
Scripts can be saved directly as .funscript files. If an OFS path is configured, you can open the result in OFS with a single click for further editing.
Built-in Video Library
A fully integrated video library lets you copy, move, delete, and edit your video files. Segment-based trimming, rotation, and direct loading into the studio are all supported from one place.
Integration Support: OFS, MPC-HC, and MFP paths can be configured for seamless launch integration — letting you jump between your tools without friction.
Windows · SR1 2.0 optimized
Reference detail link (v5.0): Detail and Video
Get it here (v6.0): Patreon link
📋 v6.0 revision notes (2026 FEB 26th)
PyInstaller Build Overhaul — Resolved cv2 module initialization failure in compiled builds. All OpenCV-dependent features (upload, ROI, tracking, thumbnails) now work reliably in the standalone .exe.
Video Delete Fix — Fixed Windows file lock error (WinError 32) when deleting videos immediately after editing. Backend now uses gc.collect + retry with backoff, and generators properly handle cleanup on disconnect.
Portrait Video Fix — Fixed crop dimensions being swapped for vertical videos (e.g. TikTok 1080×1920), which caused ffmpeg to fail or cut off parts of the frame during rotation/editing.
Subprocess Isolation — All ffmpeg/ffprobe calls now use a clean PATH environment, preventing DLL conflicts between PyInstaller’s internal libraries and system-installed tools.
Session Cleanup — New Project now fully resets all state: WebSocket connections, sync offset, playback mode, reference scripts, and segment playback cache. No more ghost data from previous sessions.
Performance Optimizations — DOM element caching during playback, batched chart rendering, rAF-throttled filter adjustments, structuredClone for undo/redo, and stack overflow prevention for 65K+ point datasets.
Security & Accessibility — XSS prevention via escapeHtml on user-supplied filenames. Aria-labels added to all icon-only buttons.
Error Resilience — Upload and frame extraction endpoints now fall back to ffprobe/ffmpeg when cv2 is unavailable. Detailed error messages replace generic “Upload failed” responses.
📋 v5.0 revision notes (2026 FEB 24th)
Smart Auto-Downscale — Automatically reduces frame resolution based on tracking mode and video size. Up to 5–6x faster on 4K videos.
Adaptive Frame Sampling — Intelligently skips redundant frames for high-FPS videos (48fps+). Up to 6x faster with Gentle preset at 120fps.
ROI-based Flow Optimization — Applies additional downscale when ROI is small relative to frame. All three optimizations stack: up to 10–15x speed gain on 4K 120fps.
Arbitrary Angle Tracking (0°–360°) — Set tracking direction freely when configuring ROI. Supports vertical, horizontal, diagonal, and any custom angle.
Dual-Pass Tracking — Non-cardinal angles are tracked via original + 90°-rotated video, combined using trigonometric decomposition. Visual axis indicator shown on ROI canvas.
Resizable Layout — New drag-to-resize splitters for graph height and left/right panel width. Ghost graph bug after “New Project” is fixed.
Device-Specific Presets — Replaced generic Gentle/Standard/Responsive with real device profiles: SR1 2.0/OSR (33ms), Handy (50ms), Basic (100ms). Point-increase bug in optimization fixed.
Full Quality Check — Client-side full-script analysis. Detects 5 issue types: impossible speed, flat regions, out-of-range values, big jumps, and stutters. Auto Fix corrects all detected issues instantly.
Precise Video Rotation — Upgraded from button-only to slider + number input + preset buttons (0°/90°/180°/270°/+90°).
Rewritten Manuals — English / Korean documentation fully updated for v5.0 with 15 new screenshots.
📋 v4.0 revision notes
Stop Tracking — Instantly abort multi-segment tracking. Interrupted segments revert to pending automatically.
Smart Segment Skipping — Already-tracked segments are skipped on restart. Only unfinished segments are processed.
Tracking Data Backup — Tracking data is auto-backed up. Deleting or modifying segments won’t erase previous results.
Chart Pan & Playhead Sync — Drag to pan the chart when zoomed in, with real-time video sync. Ctrl+drag for range selection.
ROI Precision Fix — Fixed canvas scaling mismatch causing OpenCV errors. ROI is now pixel-accurate.
Bug Fixes — Folder browse race condition fixed. Improved chart cleanup on project reset.















