Coming soon

About This Software

═══════════════════════════════════════

Huiyu AI 3D Video

Realtime AI 3D Depth Video Viewer

═══════════════════════════════════════

⚡ Turn ordinary videos into realtime 3D depth scenes you can watch, adjust, fullscreen, and experience in VR.

Huiyu AI 3D Video is not a regular video player, and it is not a “convert first, watch later” offline tool. Its core experience is realtime: play the video, estimate depth, generate depth maps, and render the 3D scene as the video is running.

During playback, AI analyzes each frame, generates depth information for the current image, and immediately reconstructs the flat video into a depth-driven 3D mesh scene with Three.js.

Foreground elements feel closer. Background elements move farther away. Ordinary 2D footage becomes a realtime 3D video space with distance, layers, and depth.

⚡ Core Feature: Realtime AI 3D Video

Huiyu AI 3D Video is designed for realtime 3D video viewing. You do not need to wait for the entire video to be processed before watching the 3D effect.

🎞️ Process while playing —— AI depth estimation runs as the video plays

🧠 Realtime depth generation —— Generate depth maps for the current frame instantly

🧊 Realtime 3D mesh —— Depth maps directly drive 3D mesh displacement

🌊 Realtime visual updates —— Video texture and depth texture refresh together

🎛️ Realtime parameter tuning —— Adjust parallax, offset, smoothing, and quality during playback

💾 Realtime cache writing —— Generated depth frames are saved locally for faster replay

🧠 AI Depth Video Reconstruction

Huiyu AI 3D Video uses AI depth estimation to analyze video frames and generate depth maps for the frame currently being played. The depth result is streamed to the frontend in realtime and used to drive 3D rendering.

This means you are not watching a static converted image. You are watching a video continuously reconstructed by AI into a realtime 3D scene.

🎞️ Per-frame depth estimation —— Generate depth information for ordinary videos

🧩 3D mesh reconstruction —— Use depth maps to create spatial displacement

🌊 Realtime depth updates —— Continuously update depth while the video plays

🖼️ Depth map preview —— View the AI-generated depth result in realtime

⚡ TensorRT-Accelerated Inference

To make realtime 3D video as smooth as possible, Huiyu AI 3D Video supports TensorRT-accelerated inference and uses a lightweight, fast depth model configuration by default.

In a compatible NVIDIA GPU environment, depth estimation can run through GPU acceleration to reduce waiting time and make video playback and depth updates closer to realtime.

🚀 TensorRT GPU acceleration —— Use the GPU for realtime depth inference

🧠 Depth-Anything-V2 —— High-quality monocular depth estimation

🌗 FP16 precision —— Balance speed, VRAM usage, and quality

🧰 Multiple model tiers —— Supports vits / vitb / vitl / vitg

🎮 Watch Video Like a Realtime 3D Scene

Huiyu AI 3D Video does more than play a video. It lets you watch the video as a realtime 3D scene. You can adjust depth amplitude, depth offset, mesh detail, edge smoothing, and visual quality parameters.

Different videos have different depth structures. You can tune the effect while watching until the scene feels natural and comfortable.

📏 Realtime depth amplitude —— Control the strength of the 3D effect

🎚️ Realtime depth offset —— Adjust the spatial baseline

🧊 Realtime mesh detail —— Balance visual detail and performance

🔄 Depth inversion —— Adapt to different depth directions

✨ Realtime edge smoothing —— Reduce depth noise and sudden jumps

🖼️ Realtime quality tuning —— Control transmission quality and playback smoothness

🥽 WebXR VR Immersive Viewing

If your browser, headset, and VR runtime support WebXR, Huiyu AI 3D Video can bring the realtime-generated 3D video scene into VR.

This is not ordinary 2D panoramic video playback. It is a 3D mesh view reconstructed in realtime from AI-generated depth maps.

🥽 WebXR VR mode —— Enter immersive 3D viewing

🧭 OpenXR runtime compatibility —— Works with environments such as SteamVR, Meta, and WMR

🖥️ Video-area fullscreen —— Focus on the realtime 3D image

🖱️ Double-click fullscreen —— Quickly switch viewing state

📂 Local Video Library and File Import

Huiyu AI 3D Video can read videos directly from a local video library and also supports manually selecting other video files. You do not need to upload private videos to the cloud; the processing workflow runs in your local environment.

📁 Local video library —— Automatically list videos from a specified directory

🎬 Open local video —— Create a playback session directly from a local path

📦 File import —— Select video files or frame-sequence archives manually

🔒 Local processing —— Videos and caches stay on your machine

💾 Realtime Generation First, Fast Reuse Later

The first time you watch a video, Huiyu AI 3D Video generates depth frames in realtime. Generated results are saved to a local cache. When you replay the same video with the same model and parameters, cached depth frames can be reused to reduce repeated inference.

In other words: the first viewing is realtime AI generation; later viewing can be accelerated by local cache.

🧠 Parameter-aware cache —— Cache is generated based on video, model, and parameters

🖼️ Per-frame depth cache —— Save generated depth results locally

🔁 Faster replay —— Avoid repeated inference

⚡ Faster over time —— Cached segments can be reused more quickly

🎛️ Professional Parameter Controls

Huiyu AI 3D Video provides multiple control parameters so you can tune the experience based on video content, computer performance, and viewing preference.

🧠 Model selection —— vits / vitb / vitl / vitg

⚡ TensorRT —— Enable or disable GPU acceleration path

🌗 FP16 —— Control half-precision inference

📏 Max depth —— Adjust depth range

🧽 Edge smoothing —— Reduce depth noise

📊 Normalization —— MinMax / Percentile clipping / Standard deviation

🎞️ Frame skipping —— Balance smoothness and full-frame playback

🖼️ JPEG and RGB quality —— Control image quality and performance

🖤 Dark Glassmorphism Interface

Huiyu AI 3D Video uses a dark gray and black glassmorphism interface. The design is quiet, clear, and suitable for long video viewing and parameter adjustment sessions.

Playback, pause, stop, VR, fullscreen, and other actions use icon-based controls. Buttons, menus, progress bars, and dropdowns all provide subtle hover feedback.

🌑 Dark gray/black visual style —— Reduce visual fatigue during long viewing sessions

🪟 Glassmorphism control panels —— Keep the interface light and layered

✨ Smooth hover animations —— Buttons, menus, and progress bars respond clearly

🎯 Icon-based controls —— Playback, pause, stop, VR, and fullscreen are easier to understand

🔒 Local-first Privacy

The core workflow of Huiyu AI 3D Video runs in your local environment. Videos come from local files or a local video library, and generated depth caches are stored locally.

💾 Local video reading

🧠 Local depth generation

🗂️ Local cache storage

🚫 No need to upload private videos to the cloud

Note: AI depth estimation and VR performance depend on your GPU, model files, TensorRT setup, browser WebXR support, and VR runtime configuration. VR mode requires a compatible browser, headset, and available OpenXR / WebXR runtime.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Huiyu AI 3D Video

Realtime AI 3D Video Architecture

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

▼ 01｜Video Source Layer

📁 Local Video Library · File Import · Video Metadata · Range Playback

▼ 02｜Realtime Depth Estimation Layer

🧠 Depth-Anything-V2 · TensorRT Runtime · FP16 Inference · OpenCV Frame Reader

▼ 03｜Depth Processing Layer

🎛️ Normalization · Percentile Clipping · Edge Smoothing · Depth Inversion · JPEG / PNG Encoding

▼ 04｜Realtime Cache Layer

💾 Parameter-aware Cache Key · Per-frame Depth PNG · Metadata Validation · Replay Acceleration

▼ 05｜Streaming Layer

🌐 FastAPI · WebSocket · Binary Frame Protocol · Local Session Management

▼ 06｜Realtime 3D Rendering Layer

🧊 Three.js · VideoTexture · DataTexture · Shader Displacement · Mesh Detail Control

▼ 07｜Viewing Experience Layer

🥽 Playback Controls · Fullscreen · Depth Preview · Glassmorphism UI · WebXR VR

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Give ordinary videos realtime depth.

Turn flat footage into a 3D space as it plays.

═══════════════════════════════════════

AI Generated Content Disclosure

The developers describe how their game uses AI Generated Content like this:

This application uses generative AI models (Depth Anything V2 / V3) to perform real-time monocular depth estimation on user-provided videos, converting 2D footage into 3D depth maps for stereoscopic and immersive viewing. All AI models are loaded and run locally on the user's own hardware. No third-party cloud AI services are used, and no user data is transmitted externally.