⚡ Quvirl Docs Hub v2.5

Quvirl Documentation & User Guides

Comprehensive guides, architecture overviews, and step-by-step instructions for the AI Video Lab and Dropshipping Product Demand Radar.

🚀 Platform Overview

Quvirl is a unified creation and growth platform engineered for video clippers, content creators, podcasters, streamers, and ecommerce marketers. It solves the two biggest bottlenecks in modern online media: finding breakout trends and product demand, and generating high-retention short-form video creative at high velocity.

💡
Clarification: Built for All Video Creators — Not Only Dropshippers: While Quvirl includes ecommerce market research tools, our AI Video Lab is an independent, studio-grade video repurposing platform. You do not need an online store or dropshipping business to use Quvirl. Thousands of video clippers, podcast editors, YouTube/Twitch streamers, educators, and social media managers use Quvirl daily to turn raw long-form videos into viral 9:16 vertical clips with dynamic captions and smart speaker framing.
🎬
AI Video Lab

Automated scene indexing, Faster-Whisper word-level transcription, high-retention animated captions, speaker tracking, and cloud rendering for 9:16 viral shorts.

📈
Product Demand Radar

Real-time demand signals combining Google Trends search momentum, Amazon sales velocity, and supplier cost structures (CJdropshipping & AliExpress).

🎯 Who Is Quvirl For?

Quvirl is purpose-built for anyone who wants to turn long-form horizontal video into high-engagement vertical short clips, as well as digital operators seeking data-backed product trends:

✂️
Video Clippers & Repurposers

Repurpose 1 to 3-hour YouTube videos, podcasts, and interviews into high-retention 9:16 vertical clips. Automatically detect conversational hooks, crop for mobile screens, and generate word-by-word animated captions with zero manual timeline slicing.

🎙️
Podcasters & Interviewers

Extract the most viral soundbites, debates, and insights from your podcast episodes. Active speaker tracking and punchy animated subtitles make your audio visual, engaging, and ready for TikTok, Instagram Reels, and YouTube Shorts.

🎮
Streamers & Content Creators

Turn Twitch streams, YouTube playthroughs, and vlog sessions into daily short-form social posts. Use your private cookie vault to import directly from YouTube without buffering or quality loss.

📦
Ecommerce Sellers & Dropshippers

Pair the Product Demand Radar (Google Trends & Amazon velocity signals) with Video Lab to turn product reviews, demos, and unboxings into high-converting social media video ads in minutes.

⚡ Quickstart Workflow

Follow this 3-step loop to find a winning product and launch viral social shorts in under 10 minutes:

  1. 1. Discover Rising Demand on Product Radar

    Open Product Radar, filter by Rising Demand Index (> 80/100), and inspect breakout Google Trends and Amazon purchase intent.

  2. 2. Import Source Video or YouTube Link

    Paste a product review URL, unboxing link, or long-form video in Video Lab. Cloud AI automatically transcribes speech and detects viral visual moments.

  3. 3. Select Viral Cut & Cloud Render

    Review curated 9:16 clips, customize caption styles and word animations, and click Render Short for instant MP4 export ready for TikTok, Shorts, and Reels.

🎬 AI Video Lab: Supported Formats & URLs

Video Lab supports direct file uploads as well as streaming cloud imports directly from top video platforms:

  • YouTube Standard URLs: https://www.youtube.com/watch?v=...
  • YouTube Shorts: https://youtube.com/shorts/...
  • Direct Media Links: Public HTTP/HTTPS links pointing to .mp4, .mov, .mkv, or .webm
  • TikTok & Vimeo: Public video share URLs
  • Local File Upload: Drag and drop videos up to 1 GB and 3 hours in length
  • Universal Stream Remuxing: Handles WebM, Opus, AAC, and multi-track audio, automatically merging them into a clean 1080p MP4 master.
  • Authorized Creator Media Import: Supports importing videos and streams that you own or hold explicit permission to repurpose.
💡
Cloud Streaming: When importing via URL, the video is fetched and processed directly on serverless cloud workers—no upload bandwidth is consumed from your local device.

⚡ Dedicated 48GB Ada Lovelace GPU & Faster-Whisper Engine

Quvirl Video Lab is powered by enterprise-grade NVIDIA Ada series Lovelace GPU (48 GB VRAM, 8 vCPUs, 64 GB RAM), delivering unconstrained throughput for real-time speech transcription and hardware-accelerated video synthesis:

🎙️
Faster-Whisper on CUDA float16

Speech transcription runs natively on dedicated GPU cores using Faster-Whisper with float16 precision. Generates millisecond-accurate word timestamps (word_timestamps=True) and eliminates background noise via integrated Voice Activity Detection (vad_filter=True).

⚡
NVENC Hardware Video Encoding

Video rendering offloads to NVIDIA NVENC silicon (H.264/HEVC hardware encoder). Produces crystal-clear 1080×1920 60fps vertical shorts with animated ASS subtitles burned in at up to 10× realtime speed.

🚀
Authorized Content Notice: Users must ensure they own the copyright or hold an explicit license for any video or stream imported into the AI Video Lab in accordance with our Terms of Service.

🤖 Primary & Secondary AI Model Stack

Quvirl Video Lab orchestrates a high-availability, multi-tier AI pipeline executing directly across dedicated NVIDIA L20 Ada Lovelace GPU cores (48GB VRAM). To guarantee uninterrupted 24/7 video processing and zero dropped jobs during high concurrent traffic, the architecture pairs every Primary AI Model with an active Secondary Fallback Model:

🎙️
Speech Recognition (ASR)
Faster-Whisper · Word Timestamps · 90+ Languages

Primary Model: CTranslate2-accelerated Faster-Whisper on CUDA float16 running directly on dedicated GPU cores. Generates sub-second transcription with word-level timestamps (word_timestamps=True) for accurate word-level subtitle synchronization and Voice Activity Detection (vad_filter=True).

Secondary Fallback: High-throughput secondary Whisper worker with automated failover, ensuring continuous transcription uptime without delay if primary GPU workers undergo high queue depth or container resets.
👁️
Vision-Language Understanding (VLM)
Nex-N2.5 Series · Keyframe Perception · Visual Indexing

Primary Model: Nex-N2.5-Pro (nex-agi/Nex-N2.5-Pro). Performs deep multi-frame visual indexing across sequential keyframes to assess visual pacing, camera dynamics, speaker framing, on-screen graphics, and product reveals.

Secondary Fallback: Nex-N2.5-mini (nex-agi/Nex-N2.5-mini) activates automatically when the primary vision model is at capacity or undergoing maintenance, providing rapid visual scene analysis with zero pipeline stalls.
🧠
Reasoning & Virality Scoring (LLM)
Qwen 27B Family · 0–100 Virality Scoring · Metadata Gen

Primary Model: Qwen 27B (Qwen/Qwen3.8-27B). Evaluates the critical first-3-second hook, emotional retention, and virality scoring (0–100), while automatically generating viral titles, descriptions, and hashtags.

Secondary Fallback: Qwen 27B High-Throughput (Qwen/Qwen3.5-27B) provides instant, seamless failover with identical JSON schema validation and prompt compatibility when the primary reasoning engine is under peak load.
📐
Smart Framing & Scene Splitting
PySceneDetect · Focal Point Trajectory · NVENC

Primary Engine: Algorithmic scene boundary cut detection combined with active focal point trajectory tracking to dynamically position the 9:16 vertical crop window centered on speakers.

Secondary Fallback: Dynamic center-weighted optical crop with automated aspect centering if complex trajectory tracking encounters extreme camera blur or low-contrast lighting.
⚡
Automated Failover Routing: Our Provider Router continuously tracks API latency, concurrent job depth, and account capacity. If a primary AI model is temporarily busy, rate-limited, or unavailable, incoming jobs automatically fail over to the secondary model within milliseconds without user intervention.

🧠 AI Scene Detection & Virality Scoring

Unlike basic video trimmers that cut at arbitrary intervals, Quvirl uses an intelligent multi-modal AI pipeline:

  • Faster-Whisper Acoustic Engine: Extracts high-accuracy speech with millisecond word timestamps and automatic punctuation across 90+ languages on CUDA float16.
  • Nex-N2.5-Pro Multi-Modal Vision Indexer: Analyzes visual pacing, camera movement, product reveals, and emotional reactions to determine natural narrative segments.
  • Qwen 27B Virality Scoring: Evaluates hook strength (first 3 seconds), speech velocity, visual dynamism, and narrative retention to give each clip a score from 0 to 100.

📱 9:16 Shorts & Viral Clips Generator

Once analysis finishes, Video Lab automatically curates the top 3–8 viral highlights and presents them in the Rendered Shorts & Viral Cuts gallery.

⚡ Hook Detection

Pinpoints compelling opening statements or visual surprises to maximize TikTok/Reels scroll-stopping power.

🎯 Speaker Crop

Intelligently frames speakers and product demonstrations in vertical 9:16 aspect ratio without distortion.

⏱️ Custom Trimming

Adjust start and end timestamps precisely using interactive scrubbers before sending to the cloud render engine.

🔥 Dynamic Kinetic Captions & Styling

Quvirl features high-impact, word-level animated subtitles engineered for high-retention mobile viewing:

  • Word-by-Word Highlight: Currently spoken words illuminate in vibrant colors (yellow, green, electric blue) with optional bounce animation.
  • Font Customization: Select from bold, high-legibility display typefaces optimized for mobile screens.
  • Vertical Position Controls: Place subtitles at bottom, middle, or lower-third to avoid overlapping TikTok / Reels native UI overlays.
  • Uppercase Punch: Automatic capitalization and emoji insertion for high-engagement keywords.

⚡ NVENC Cloud Rendering Pipeline

When you click Render Short, the clip is synthesized in the cloud using an optimized NVIDIA NVENC hardware-accelerated pipeline:

  1. Hardware Crop & Scale: Dedicated NVENC silicon crops and scales to crisp 1080×1920 (9:16) with smart speaker framing.
  2. ASS Animated Subtitle Burn-In: High-contrast, anti-aliased styled dynamic captions burned into the video stream with zero dropped frames.
  3. Audio Mastering: AAC 192kbps stereo normalization and loudness leveling for mobile speakers.
  4. Instant MP4 Social Export: Generates a direct MP4 download link ready for immediate publish to TikTok, Instagram Reels, and YouTube Shorts.

🍪 Cookies Vault: Why Cookies Are Needed

Platforms like YouTube frequently enforce strict bot-detection mechanisms (such as "Sign in to confirm you're not a bot" or age restrictions) on popular or music videos. By providing authentication cookies, Quvirl's cloud downloader can authenticate requests just like a legitimate browser, allowing 100% reliable video imports.

🔒
Strict Privacy & Account Isolation: Cookies uploaded to Quvirl are stored exclusively in your private user session vault. Other users, incognito visitors, and public workers cannot see or access your cookies.

📥 How to Export Netscape cookies.txt

To export your cookies safely from your desktop browser:

  1. Install a Netscape Cookies Extension

    Install a trusted Chrome or Firefox extension such as "Get cookies.txt locally" or "Cookie-Editor" from the official Web Store.

  2. Navigate to YouTube

    Go to youtube.com while logged into your account to make sure active session cookies are present.

  3. Export as Netscape format

    Click the extension icon and select Export as Netscape Format (.txt). Save the file to your computer.

📤 Uploading to Video Lab Vault

In Video Lab, click the 🍪 Cookies button in the header actions. Drag and drop your cookies.txt file into the modal or paste the text directly. The system immediately parses, validates, and stores your credentials securely in your isolated session vault.

🔍 Live Cookie Testing Tool

Before launching video imports, Video Lab provides an automated pre-flight testing tool to guarantee your YouTube authentication is active and unrestricted:

  1. Automated Token Inspection

    Upon upload, the system automatically checks your cookie file for active Google / YouTube authentication keys, including LOGIN_INFO, SAPISID, __Secure-3PAPISID, SID, and SSID.

  2. Real-Time Cloud Verification

    Clicking 🔍 Test YouTube dispatches a lightweight probe directly from our dedicated GPU worker to YouTube using your credentials. It validates that your session authenticates your authorized creator session cleanly.

  3. Resilient Multi-Client Fallback

    During video downloads, our importer automatically queries alternative player clients (including TV, Web-Embedded, and MWeb) to ensure stable playback compatibility for authorized creator media.

🛡️
Session Vault Isolation: Every user's cookies are partitioned strictly into an isolated directory keyed to their private session (usr_<id>). Requests are proxied securely through the Vercel serverless gateway (/api/worker) with full encryption in transit.

📈 Product Demand Radar Overview

The Product Demand Radar monitors search engines, social channels, and marketplace buying behavior to uncover breakout dropshipping products before competitor saturation.

🛒 Amazon Demand Validation

Validates whether search curiosity translates into actual purchasing intent by tracking:

  • Monthly sales unit estimates
  • Best-Seller Rank (BSR) delta across subcategories
  • Review velocity (number of new verified buyer reviews per week)

📦 Supplier Economics & Profit Margins

Automatically pairs validated demand keywords with real product listings from CJdropshipping and AliExpress:

  • Factory Price: Baseline unit cost from verified suppliers
  • Shipping Cost: Standard air express rates to primary markets (US, UK, CA, AU)
  • Recommended Retail Price (RRP): Based on competitor market averages
  • Net Profit Margin: Estimated profit per unit after product and shipping costs

⭐ Rising Demand Index (0–100)

Every product is ranked by Quvirl's composite scoring algorithm:

// Quvirl Demand Index Formula
Score = (Trends_Velocity × 0.40) + (Amazon_Purchase_Intent × 0.35) + (Margin_Viability × 0.25)

Scores above 75 represent strong breakout candidates with healthy margins and growing consumer interest.

🛡️ 10-Minute Video Retention Window

To sustain high platform throughput (supporting 1,600+ videos per day without disk exhaustion), Quvirl implements an automated lifecycle for source media files:

⏱️
Strict Post-Ready Timer: The 10-minute countdown starts strictly after a video is marked ready. Downloading, transcribing, and visual indexing time is never deducted from your viewing and clipping window.

After the 10-minute window expires, the large raw video file (source.mp4) is safely pruned to reclaim disk space. All AI analysis, transcripts, viral clips, and rendered shorts remain permanently intact in your dashboard.

🗑️ Immediate Failure Cleanup

If an import or analysis fails due to an invalid URL, unsupported codec, or user cancellation, all temporary and source video files are deleted immediately from the cloud container to preserve maximum capacity for other users.

⚖️ Fair-Share Concurrency Limiting

Each user session is permitted up to 2 active concurrent processing jobs. This ensures zero wait times and prevents any single automated script from monopolizing cloud GPU and CPU rendering capacity.

📏 1GB File Cap & 3-Hour Duration Guardrails

To protect platform stability, the following upfront constraints are enforced:

  • File Size Cap: Maximum file upload size of 1,024 MB (1 GB). Requests exceeding this are rejected upfront with HTTP 413.
  • Video Duration Cap: Maximum video duration of 3 hours (10,800 seconds).
  • SSRF Defense: All import URLs are validated against private IP ranges (127.0.0.1, 10.0.0.0/8, 192.168.0.0/16) and cloud provider metadata endpoints (169.254.169.254).