Every pillar, expanded.

Channel Assistant generates the brand bible. Per-scene image prompts. Hook lab. Retention-aware structure.

• Channel Assistant: paste a 1-line niche, get a brand bible (tone, banned topics, recurring motifs) + 50 video ideas in a single call (~$0.08 per channel).
• Per-scene generation: each scene gets a paragraph script + an image prompt tuned to the brand. Edit one without regenerating the others.
• Hook Lab: generate 5+ opening hook variants per video, then track which performs in real channel data.
• Retention-aware structure: scripts include markers for the 30-sec retention check, the mid-video re-hook, and the call-to-action.
• Profit Mode caps AI spend per video — set a $0.50 ceiling and the orchestrator stops when you hit it.

Ken Burns, parallax, tilt-shift, color LUTs, beat-aligned cuts, speed ramps, PSD layers, keyframe timeline — all rendered locally.

• 8 bundled motion presets — Ken Burns, parallax, tilt-shift, dolly zoom, speed ramps and more — applied per-scene via dropdown.
• 6 bundled color LUTs (.cube) + custom LUT upload. Each scene picks its own or inherits the channel default.
• Beat-aligned cuts: drop a music track and the engine detects beats and snaps scene boundaries to them.
• PSD layer import: layered Photoshop files split into independent transformable layers (auto background removal).
• Keyframe timeline: drag dots for opacity/scale/rotation/position over time, with bezier curves and snap-to-beat.
• All renders run via FFmpeg on your local GPU, with bit-identical regression tests guaranteeing stable output.

A per-scene asset studio: pick, mix and sequence clips and images, preview every clip inline, and fit it all to your narration.

• Playable mini-previews: every generated AI clip plays right in the Visuals page so you can judge a take before it renders.
• Pick & mix per scene: toggle the image and each clip independently, combine stills and motion, and drag to reorder.
• Multiple clips per scene: each clip is its own card with a per-clip delete, auto-stitched in order.
• Fit to narration: selected assets are sized to each scene's voiceover, with one-click 'Reset to auto'.
• Fully additive — leave a scene untouched and it renders exactly as before; the bit-identical render path is unchanged.

Auto-detects and benchmarks your hardware, then renders on the fastest GPU encoder available — NVENC, QSV or AMF — with a safe CPU fallback.

• Hardware detection: probes NVIDIA / Intel / AMD GPUs, CPU, RAM, OS and FFmpeg hwaccels to learn what your machine can do.
• Real test-encodes: every candidate encoder is validated with a 1-frame render and marked Verified or Failed — never trusted blindly.
• Benchmark + auto-select the fastest stable encoder (NVENC → QSV → AMF → libx264) with a guaranteed CPU floor.
• Render-time fail-open: if a GPU encode misbehaves mid-render, it falls back to the byte-identical libx264 command.
• Render Diagnostics panel + manual encoder picker — see what each device detected and pin any verified encoder.

Auto-track thumbnail rotations, title changes and hook variants, with Z-test significance, AI recaps and per-video Improve suggestions.

• Auto thumbnail-rotation tracking + a 6-hour title poller that catches title edits.
• Z-test statistical significance flags the winner once the sample and effect size are big enough.
• Auto AI recap (~$0.02/finalized experiment): a plain-English write-up of what worked and why.
• Per-video Improve panel: ingests retention drop-off + experiments + analytics and ranks 3-8 evidence-backed fixes by severity.
• Sortable Experiment Journal across all channels — pivot by variant type or win-rate.

YPP progress, recent trends, render failures and dormancy alerts — the Command Center surfaces what needs attention today, across all channels.

• YPP progress tracker: real subscriber + watch-hour counts with the Partner Program thresholds overlaid.
• Trend charts: subscriber growth, watch hours and revenue for the last 30/90/365 days per channel.
• Render-failure + dormancy alerts: channels idle 14+ days surface with severity-ranked recommendations.
• Command Center: a cross-channel dashboard answering 'what needs my attention right now?' with a priority queue.
• Pure read-aggregation — no extra AI calls on page load.