Architecture & Operational Blueprint

How ProPeek Reverse-Engineers Any Media

The complete technical and creative walkthrough on turning raw videos, images, and concepts into production-grade prompts and stock SEO metadata — 100% inside your browser.

ProPeek 3-Stage Architecture Diagram
Sequential 3-Stage Client-Side Pipeline
• 0 MB Server Upload• Direct Provider Inference• Agency-Compliant CSV
01
Stage 1: Media Ingestion

Client-Side Keyframe Sampling & Zero-Server Upload

Traditional AI video analysis tools upload entire multi-gigabyte video files to cloud servers, introducing massive delays, privacy vulnerabilities, and bandwidth costs. ProPeek completely revolutionizes this with an entirely in-browser pipeline:

HTML5 Video Canvas Decoding: When you drag and drop an MP4, MOV, or WebM file, your browser renders frames locally to an off-screen canvas element at designated temporal intervals.
Adaptive Density Control: Keyframes are intelligently sampled from the start, midpoint, and climax of the clip to capture camera acceleration, lighting shifts, and subject actions.
Zero Video Data Transferred: Only compressed lightweight frame snapshots are passed to the AI provider. Your source footage never leaves your personal device.
100% Client-Side Privacy Instant Zero-Wait Load MP4, MOV, WebM, JPG, PNG
02
Stage 2: Vision AI Analysis

Multi-Layer Cinematography & Semantic Deconstruction

The extracted keyframes are sent directly to your configured Vision AI provider (Google Gemini 2.0 Flash / Pro or Groq LLaMA 3.2 Vision) using multi-layer structured system prompts. ProPeek deconstructs the scene across five critical visual pillars:

Vision AI Stock Metadata and Prompt Analysis Mockup

Camera Kinematics & Optics

Identifies focal length (e.g. 24mm anamorphic vs 85mm prime), depth-of-field, aperture bokeh, and camera movement trajectory (orbit, pedestal, crane, dolly, whip pan).

Lighting & Atmospheric Depth

Detects volumetric lighting, key/fill ratios, rim highlights, color temperature (warm golden hour vs cold cyan sci-fi), haze, mist, and shadow contrast.

Subject & Temporal Progression

Captures subject anatomy, wardrobe textures, facial expressions, and temporal changes between the start and end of the shot for seamless video prompt generation.

Render Style & Medium

Classifies medium accurately: 35mm celluloid film, digital cinema RAW, 3D Unreal Engine 5 render, isometric vector, or hyper-realistic commercial photography.

03
Stage 3: Synthesis Engines

Tailored Generation Modes for Every Creator Workflow

Depending on whether you want to recreate footage, create original stock assets, or batch-tag existing files, ProPeek offers 3 specialized modes:

A. Exact Style Reverse-Engineering

Generates an exhaustive technical prompt matching the exact reference clip or image. Ideal for inputting into Midjourney v6, Black Forest Flux, Google Veo 3, ByteDance Seedance 2.0, and Kling to achieve visual fidelity.

B. Inspire Mode (Stock Agency Safe)

Extracts the dynamic camera motion, lighting palette, and artistic framing of the reference, but invents a completely distinct subject and environment. Ensures 100% originality and zero copyright complications on stock agencies.

C. High-Speed Metadata Only Mode

Designed for contributors who already have finished footage and photos. Bypasses the prompt reverse-engineering step, cutting API token usage in half and completing batch analysis in under 3 seconds per file.

04
Stage 4: Agency Compliance

Commercial Sanitization, Deduplication & 1-Click CSV Export

Submitting stock media to Adobe Stock, Shutterstock, or Freepik often leads to rejected files due to trademarked words, duplicate tags, or improper CSV headers. ProPeek automates rigorous compliance:

Automatic Trademark Scrubbing: Filters out brand names, camera manufacturers (Sony, Canon, RED), artist names, and copyrighted terms that trigger marketplace rejections.
50-Keyword Limit & Relevance Sorting: Deduplicates keywords, orders them by search volume relevance, and strictly caps them to agency guidelines.
Taxonomy Mapping: Maps each asset directly to official Adobe Stock (21 categories) and Shutterstock (26 categories) numeric IDs.
Instant CSV Generation: Export ready-to-upload spreadsheet files with proper UTF-8 formatting and agency column headers in 1 click.
Expert Tips

Getting Maximum Accuracy From ProPeek

  • •Gemini 2.0 Flash: Best for high-speed multi-frame video analysis and complex lighting identification.
  • •Prompt Lab Variations: Use the Prompt Lab to generate 5 variations of your master prompt with varying degrees of creative divergence.
  • •Batch Processing: In Metadata Only mode, you can drop dozens of assets sequentially and download a single consolidated CSV for your entire submission.

Ready to Reverse-Engineer Your Media?

Drop any video or image into the workspace and generate production prompts and SEO metadata in seconds.

Launch Free Workspace