Video Skills & Edits Knowledge Base

Contains zero-dependency FFmpeg command-line primitives for Ken Burns zooms (in/out, pan left/right), zoom punch (~0.5s digital burst), 9:16 safe-zone fit with blurred background, LUT-style color grading presets (warm, cool, cinematic, contrast), post-effects (film grain, vignette, light leak, minterpolate motion blur), xfade transition types, and ASS/SRT caption burn-in.

A complete 5-stage Python automation script requiring zero external pip dependencies beyond FFmpeg. Executes Stage 1 (Image to Video), Stage 2 (Concatenation with xfade), Stage 3 (A/V Audio Muxing with apad + shortest), Stage 4 (Caption Burn-in), and Stage 5 (QA Gate check).

Pre-configured parameter blueprints for specific video archetypes: Motivational Reels (8 images + voiceover), Listicle Videos (clean transitions + subtitle), Silent Slideshows, and Quick Single-Image Videos.

An assert-based bash script verifying output file existence, resolution (1080x1920), frame rate (≈30 fps), stream validity, and non-zero duration.

Categorizes core video editing effects (Visual Effects, Motion Graphics, Color Grading, Sound Design) and outlines ideal application cases for maximum audience retention.

Establishes non-negotiable retention principles (0-3s hook rules, dead-air elimination, pattern interrupts) and provides a tailored hybrid stack combining screen recordings with cinematic overlays.

Individual brand positioning, visual identity systems, recording workflows, short-form vs long-form adaptations, agent prompts, and mistake checklists for 4 distinct SaaS products.

Synthesizes visual effect categories with retention optimization rules specifically calibrated for app-screen content.

Mandatory rules for all application screen recordings: no raw mouse pointers, mandatory 1.1x zoom punches on calculated values, and SDR tone-mapping.

Includes complete 8-part sub-specs for Dose Calculator / Clinical Core, Female ProMedic, Coach ProMedic, and Dentist Pro App. Covers exact hex color codes, font hierarchies, screen recording post-processing, TikTok/Reels retention hybrids, YouTube long-form adaptations, growth notes, and copy-paste prompt templates.

Bash deployment script for installing Homebrew CLI dependencies (ffmpeg, ffprobe, opus-tools), Python isolated venv, pinned packages, Whisper models, and verification steps.

State management, VFR-safe command execution runner, proper FFmpeg escaping, filesystem helpers, and platform delivery budget enforcement (bitrate/size caps).

vision_bridge.py API key rotation for cloud vision models + perceive.py local OpenCV color variance & Haar-cascade face detection.

Beat-to-asset semantic matching (Ad mode 5 beats / Narrative mode 7 beats), motion path calculation, VFR-safe rendering, xfade duration alignment, and A/V drift correction (apad + -shortest).

Strict 3-state (PASS / FAIL / INCONCLUSIVE) verification layer (qa_gate.py & verification_engine.py). INCONCLUSIVE or FAIL strictly blocks "done" status.

visual_judge.py visual quality scoring, Brain orchestrator 3-pass loop, theme seed auto-capture, VLM critique, and EBU R128 loudness normalization (loudnorm=I=-14:LRA=11:TP=-1).

Defines the exact sequence agents must follow: 1. Hook (0-3s), 2. Pacing/Dead-air removal, 3. Kinetic Captions, 4. Audio Engineering, 5. Seamless Loop, 6. Visual Effects.

Detailed execution steps from Step 0 (Ingest & Analysis) to Step 7 (Mandatory Self-Evaluation before Output). Hard-codes against failure modes like A/V sync drift and unreadable text.

Provides copy-pasteable AGENTS.md and system prompts for configuring LLMs to act as short-form viral video optimization specialists.

Protocol for clean screen capture: minimum 1080x1920 (prefer 1284x2778 for room to zoom), SDR tone-mapping for HDR recordings, 10% edge margin rule for UI elements, and mandatory raw archiving.

Step 1: Color Grade preset (clean high-key, 5600K white balance, +0.2 EV, black lift RGB 8-10, highlight cap RGB 245, saturation -5% to -8%). Step 2: Motion (1.10x-1.15x zoom punch targeted specifically on result/calculation numbers).

Deep dive into 2026 platform algorithms (TikTok, Reels, Shorts, LinkedIn). Details engagement hierarchy (Re-watch > Share > Comment > Like) and the critical 1.3s visual hook threshold.

Catalog of proven hook formulas vs hook killers, visual movement pacing, sound design rules, and platform-specific format parameters.

Systematic troubleshooting guide for diagnosing underperforming videos (Low Retention = Weak Hook; High Drop-off at 0:05 = Slow Pacing; Low Shares = Weak CTA/Relatability). Includes the Mute Test and 2026 Viral Verification Checklist.

Configuration for running headless 2K H.265 (HEVC) screen recording pipelines on Hetzner Dokploy Linux server (1440x2560 pixels, libx265, -crf 18, -preset fast, strict 50.0s duration).

Maps out 25 generated videos across 5 web apps (clinical, dentist, coach, female, ielts). Outlines human-like automated browser dynamics: Bezier mouse paths, slow scrolling, feature clicking, and simulated user inputs.

Includes a complete Python script wrapping faster-whisper, OpenCV, and FFmpeg for auto-generating subtitles, color analysis, and video assembly.

Analyzes technical limits of Python in video pipelines: complexity explosion when chaining filters, pure code vs AI intelligence, FFmpeg process locking, and memory bounds.

Pre-authorized SSH connection strings for remote Linux servers (hostinger, contabo2_new1, hetzner_dokploy, ai_developer_key).

Art style mappings for video generations: Type-C Operational, Type-4 90s Comics, Type-5 Heavy Comic Gutters, Type-7 EPO3, Type-6 Beige Vintage Editorial, Style-10 Watercolor Detective, Style-13 Detective Sketch Collage.

Maps exact file locations for app screen pictures, episode prompt files, and source content for Coach, Dentist, IELTS, Female ProMedic, Doctor-X, and Promedic1 core.