Funscript generator for Windows
AutoScript Sync is a funscript generator for Windows PCs, sold for Windows 10 and 11, 64-bit. It installs for your user account without admin rights from an installer of about 2 GB that carries its own AI runtime, so you do not need Python. An NVIDIA graphics card speeds up the AI through CUDA; without one it runs on the CPU, more slowly.
Checked against the app's code on 21 September 2026. Describes version 2.121.1, the current release.
Which Windows versions
We sell it for Windows 10 and Windows 11, 64-bit. To be exact about what that rests on: the installer sets no minimum Windows version and does not check the architecture, the app runs no version check, and the development notes mention only Windows 11. We have no Windows 10 test record. The one hard 64-bit dependency is VLC, which the installer fetches in its 64-bit form. If you are on Windows 10, run the free trial before buying.
Published minimums for RAM, graphics memory (VRAM), CPU or free disk space do not exist, because none has been measured. We would rather say that than print numbers nobody checked. The installed app is larger than the 2 GB installer, since the bundled AI libraries unpack to several gigabytes.
What the installer does
- Installs for you only
- Per-user install with no admin rights, into
%LOCALAPPDATA%\Programs\AutoScriptSync. - Brings its own runtime
- About 2 GB, the same file serves as the update download. It bundles the full PyTorch, torchvision and Ultralytics stack with CUDA support, plus ONNX Runtime, so no Python install is needed.
- Installs VLC if missing
- The app plays video through VLC. If the installer finds no VLC in the machine-wide registry, it downloads the 64-bit VLC installer from videolan.org and runs it silently.
- Bundles the device engine
- A headless Intiface engine ships with the app and starts on demand for Intiface devices, so you do not need Intiface Central installed.
- Downloads the AI models
- An optional installer task, or the first launch, downloads YOLOv8 small-pose and MiDaS small (about 109 MB together) into your user folder. The installer's wording says the models are sized to your hardware; in fact it always fetches these two, and the in-app Download best button is where you choose larger ones.
The installer also offers a step that would fetch a hardware-matched AI runtime. Those runtime packs are not published, so the step is skipped and the app uses the runtime it already bundles; GPU support comes from the bundled CUDA libraries.
Graphics card: NVIDIA CUDA, or the CPU
No graphics card is required. The Hardware setting offers Auto (the default), GPU (CUDA) and CPU.
| Hardware | AI models (pose, depth) |
|---|---|
| NVIDIA card with CUDA | Accelerated through CUDA |
| AMD or Intel graphics | No AI acceleration: runs on the CPU |
| No graphics card | Runs on the CPU, much more slowly |
Only the neural models use the GPU. Optical flow always runs on the CPU, and video decoding uses any vendor's hardware decoder where one is available. If CUDA is asked for but cannot be used, the models fall back to the CPU automatically and the app reports the provider actually in use, so you can see when that has happened. TensorRT is not used.
To see where your analysis runs, press Check GPU (also View → Check GPU…). It reports whether PyTorch and ONNX Runtime can see the GPU, the VRAM in use and, with a pose model installed, the active pose provider and a speed self-test in milliseconds per frame.
How long a conversion takes: the one measurement we have
On the developer's PC with an NVIDIA RTX 5080, at Balanced, films of 32 to 39 minutes took between 8.2 and 28.5 minutes to generate.
That is one GPU, one setting and a few films. Nothing has been timed on other graphics cards or on the CPU alone, so we cannot tell you how long your PC will take. The trial will.
For a sense of the GPU-to-CPU gap: a code comment records the small depth model at about 26 ms per frame on that RTX 5080 against about 234 ms on the CPU. That is one model, not a whole run.
Where your data lives
%LOCALAPPDATA%\Programs\AutoScriptSync- The program itself.
%USERPROFILE%\.autoscript_sync- Your user data, downloaded models, everything you have taught the AI (including
corpus_learner.jsonand the AI Bulk Learn review queue), and thelogsfolder with the latest 40 generation reports. %USERPROFILE%\.autoscriptsync- Your licence. Note the different spelling: no underscore.
- Beside each video
- The generated
.funscript, its.bakbackup, and any.roitrack.jsonfrom Live Track.
Uninstalling removes the program but leaves your user data in place, so months of teaching survive an uninstall and reinstall. Delete .autoscript_sync yourself if you want it gone. To move to another PC, free a seat first with Help → License / Activation… → Deactivate this PC.
Updates
At each launch the app checks autoscriptsync.com for a new version and shows a notice if there is one. It never installs an update by itself. In practice you update by downloading the new installer from the download page and running it over the old one; your data and licence stay. Help → Check for Updates… can fetch and run the installer only once a separate download code has been entered, which the licence email does not include.
Installed AI models are checked for newer versions at launch and re-downloaded when the published file changes. The bundled device engine is not updated this way. Your licence is lifetime and not tied to a version, so the same key activates every release.
If something is wrong
- Generation is much slower than expected. The models are probably on the CPU. Press Check GPU. If you have an NVIDIA card and it is not used, set Hardware to GPU (CUDA) and read the provider the app reports.
- The app closes straight after launch with no message. VLC is most likely missing: if the installer could not download it, the app currently exits without showing why. Install 64-bit VLC from videolan.org and start the app again.
- Models did not download. The first-run download needs internet. Use Download best in the app, or check that automatic model downloads are on in the AI engine settings.
Still stuck? Help → Export Diagnostics zips your reports, which record file names but no folder paths or video, to send to support. Funscript troubleshooting covers script and playback problems.
Questions
Will it use my AMD Radeon or Intel Arc graphics?
Not for the AI models, which run on the CPU on those cards. Video decoding may still use their hardware decoder.
How much VRAM do I need?
We have not measured it, so we do not publish a figure. The app caps the depth model's GPU memory pool at 4 GB and frees cached GPU memory between videos in a batch. Check GPU shows the VRAM in use on your card.
Does it need the internet?
For activation, starting the trial, the first model download and the Autoblow AI Ultra's cloud, yes. Generating scripts works offline, and an activated licence keeps working if the licence server cannot be reached.
Read next
- What the funscript generator doesThe product: Generate, Accuracy, editing, devices, price.
- The AI models and where they runPose, depth, flow and the classifier, part by part.
- Devices through IntifaceHow the bundled engine connects strokers.
- Download AutoScript SyncThe installer for the trial and for buyers.
Check it on your own PC first
The trial is the same installer buyers get, with every feature unlocked for one day; a licence is £14.99 once.