🌙
Grayscale depth map
Official Video-Depth-Anything (Small, Apache-2.0) with temporal consistency, so brightness stays stable across frames. It is widely used as the depth control signal that downstream creative tools expect.
Windows 10 / 11 · Runs fully offline · GPU accelerated
Turn any video into a
Turn any video into a
depth map & skeleton
Drop in a clip, pick a mode, hit convert. XIANGYU Deep Video Converter generates grayscale depth maps, human pose skeletons and 478-point face point clouds — all on your own machine. The original audio is preserved automatically, and nothing is ever uploaded.
Drop your video here
sample.mp4 · 00:12 · 1920x1080
🌙Depth map
🦵Pose skeleton
🧬Depth + pose
🧑Face 478 points
🎦Stacked all
🎨Color depth
Depth inference · GPU68%
Audio preserved · AAC 192k00:08 / 00:12
Six conversion modes
One clip, six looks. Intermediate depth and pose results are cached, so switching modes does not re-run inference.
🦵
Human pose skeleton
PyTorch KeypointRCNN (COCO 17 keypoints) draws the skeleton over the original footage, making motion trajectories readable at a glance.
🧬
Depth + pose stacked
Depth map as the base layer with the pose skeleton stacked on top — spatial layering plus visible motion.
🧑
478-point face cloud
MediaPipe FaceLandmarker extracts 478 facial landmarks rendered as a colored point cloud, preserving every expression detail.
🎦
Stacked all
Depth base + pose skeleton + face point cloud in a single overlay, for maximum information density.
🎨
Color depth
KMeans analyzes dominant colors and maps them to pseudo-depth (black is always 0), or supply your own color-map JSON to control each color precisely.
Nothing leaves your machine
This is not a cloud service. There is no server to upload to, because there is no upload code.
🔒
No upload, ever
The interface is a local web service bound to 127.0.0.1. Your footage never crosses the network boundary, which also means no queue, no rate limit, no credit system.
⚡
Your GPU does the work
Inference runs locally with CUDA when a compatible GPU is present, and falls back to CPU automatically otherwise. Each completed job tells you which one it used.
🎤
Audio preserved
The original soundtrack is remuxed into the output at 96k-256k (192k default), with channels configurable. Start time and duration trims are applied to audio in sync.
📦
Models bundled
Every model ships inside the installer. After installation you can disconnect from the internet entirely and everything still works.
🦠
No account needed
No sign-up, no cloud project, no telemetry dashboard. You install it, paste your license key once, and use it.
💰
Pay once
A single lifetime license, no subscription and no seat fee. Verification is offline Ed25519, so licensing never depends on our servers staying up.
Scope of the tool: XIANGYU processes video files you already own — it does not
generate synthetic stills or clips from text prompts, and it does not swap, animate, clone or
otherwise transfer a person's identity onto another face. The face landmark mode simply draws
478 detected landmarks as a colored point cloud over your own footage, the same way the pose
mode draws a skeleton. Outputs are derivative visualisations of your input video, rendered
locally and never uploaded.
One-time purchase. No subscription.
The trial has every feature enabled — only clip length is capped at 10 seconds. Run your own footage first, then decide.
Lifetime
Full license
$9.99
One payment · yours permanently
- Lifetime license for one machine, no subscription
- Removes the 10-second per-clip trial cap
- Includes all 1.8.x updates
- License key delivered automatically after payment
Delivery: immediately after payment you receive the license key and the installer download link by email.
Paste the key into License inside the app — verification is offline Ed25519, so no internet and no machine code are required.
Payments for international orders are processed by our merchant of record, which handles VAT / sales tax and invoices.
See the refund policy (7-day money-back guarantee).
FAQ
Does it require an internet connection?
No. Model weights ship with the installer, so conversion works fully offline. The network is only used when you deliberately open the payment page.
Can I use it without a discrete GPU?
Yes. If no usable CUDA device is detected the app falls back to CPU automatically. All features remain available, only slower, and each job reports whether it ran on GPU or CPU.
Does the exported video still have sound?
Yes. The original audio track is remuxed automatically at your chosen bitrate (96k-256k, 192k by default). Trimming by start time or duration trims the audio track in sync.
Is any of my footage uploaded to a server?
Never. The app is a local web service bound to 127.0.0.1; all processing happens inside your own machine and there is no upload logic anywhere in the codebase.
Which video formats are supported?
Anything ffmpeg can decode (mp4, mov, mkv, avi, webm and more). Output is available in libx264, libx265 and libvpx-vp9.
What are the installation requirements?
Windows 10 or Windows 11 64-bit, roughly 6 GB of free space (D:\XIANGYU by default, with automatic fallback). Double-click the desktop shortcut afterwards — no need to install Python or ffmpeg yourself.
Is the interface available in English?
Version 1.8.7 — the current download — ships with a Chinese interface. A full English interface (application, installer and launcher) is in development for version 1.9.0 and will be a free update for every licence holder, so buying once now means you receive it at no extra cost.
How do refunds work?
There is a 7-day money-back guarantee. Email us if the software does not work for you and we refund in full within 7 days of purchase. Full terms are on the refund policy page.
Windows x64 · current version v1.8.7
Download XIANGYU Deep Video Converter
Windows x64 installer, about 2.6 GB including all model weights, with resume support. Everything works offline after installation.
Support
Questions? Email the developer directly
This is an independent, fully local tool with no ads, no telemetry and no data monetisation. Installation, licensing and conversion results — any question is welcome by email, and replies usually arrive within 24 hours.
Email: luyingzhi2@gmail.com
Buy license · $9.99 lifetime