PRODUCT: AI-powered smart video cutter / clip assistant named "Cutter". CORE FLOW:…
- User
PRODUCT: AI-powered smart video cutter / clip assistant named "Cutter". CORE FLOW: Upload video → AI analyzes → Chat with AI to find moments → Timeline shows cuts → Download selected clips OR remaining parts. KEY FEATURES: drag-drop upload, chat with quick prompts, interactive timeline with cut segments, play selected cuts, long-video segments, download selected/remaining, status: uploading | analyzing | ready | processing | done. NO backend required for mock — use realistic mock data and UI states.
Design a full-screen AI video editing web app with a cinematic, dark, Netflix-meets-Final-Cut aesthetic.
Layout:
- Left: collapsible chat assistant "Clip AI" with message bubbles, suggested prompt chips, typing indicator
- Center: large video player with custom controls and a rich multi-track style timeline underneath showing AI-selected green cut ranges; clicking a cut plays only that range
- Right: project panel — video metadata, duration, token/usage meter, segment cards for long videos, download actions
States to design: empty landing with dramatic dropzone, uploading progress, analyzing with AI pulse animation, ready workspace, processing export, success toast.
Typography: sharp, modern. Accents: electric violet + cyan on near-black. Glass cards, subtle grain. Mobile: stacked tabs Chat | Player | Export.
Include quick commands: "Find funniest moments", "Keep only talking parts", "Skip silence", "Video summary", "Most exciting 5 moments". Make it feel premium and creative, not enterprise. Constraints: responsive (desktop first, workable mobile); real UI copy not lorem; realistic timestamps HH:MM:SS; loading and empty states; accessible contrast; mock only (no real FFmpeg/Gemini); clean React-ready component structure.
Assistant
I'll build "Cutter" — a cinematic AI video clip assistant. Let me start by exploring the project structure.
Now I'll generate cinematic frame assets and set up the theme foundation.
Now the data layer — types, formatters, and rich mock content.
Now the state engine — a context provider with the full status machine and simulated playback.
Now the icon set and shared UI primitives.
Now the dramatic landing / upload / analyzing experience.
Now the Clip AI chat panel with collapsible rail, typing indicator, and quick commands.
Now the cinematic player with custom controls.
Now the centerpiece — the multi-track timeline with cut ranges, waveform, markers, and minimap.
Now the right-hand project panel with metadata, usage meter, segments, and export flow.
Now the toast, top bar, and the app shell that ties it all together.
Let me build to verify everything compiles.
A few polish passes — copy accuracy and contrast on the smallest text.
Cutter is built — a full-screen, cinematic AI clip assistant with a complete mock pipeline.
What's there
Landing → Upload → Analyze (
Landing.tsx)- Dramatic dropzone with real drag-drop, drifting violet/cyan glows, grain, scan-line hover sweep, and "Use the demo episode" shortcut
- Uploading: live % , GB counter, chunk/region/encryption stats, shimmering progress bar
- Analyzing: triple pulse-ring AI core + 8-step checklist ("Transcribing 12,486 words", "Detecting scenes & 47 silent gaps") with per-step timings
Workspace
- Left — Clip AI chat: collapsible to an icon rail, message bubbles with markdown-ish rendering, credit/meta footers, bob-animated typing indicator ("scanning 01:12:38 of footage…"), and the 5 quick commands (funniest moments, only talking parts, skip silence, video summary, top 5 exciting). Each command rewrites the timeline with a distinct realistic cut set and a written rationale; free-text gets keyword-matched replies (shorts/9:16, pull-quotes, intro, silence).
- Center — player: cinematic frame that swaps by chapter with Ken Burns while playing, burned-in timecode, PLAYING/PAUSED tally light, scanlines + grain, active-cut banner, scrub bar with green cut inlays, ±10s jog, VU meter, loop-cut toggle, play-selected, 0.5–2× speeds.
- Center — timeline: minimap with viewport window, HH:MM:SS ruler with chapter boundaries, V1 filmstrip track with green cut ranges (click = play only that range, hover = AI reasoning + confidence, checkbox = include/exclude, ✕ = delete), A1 waveform colored by kept/removed with −42 dB gate label, AI event track (laughs, applause, keywords, silence) with tooltips, draggable playhead, 1–8× zoom with auto-follow.
- Right — project panel: poster + metadata grid, kept/removed split with live percentages, AI credit meter, six long-video segment cards with per-chapter cut counts and jump/play, format + quality pickers, and both export actions.
States:
uploading → analyzing → ready → processing → done— export shows a staged render banner ("Cutting on keyframes", "Normalising to −14 LUFS"), then a file list with realistic names/sizes, zip action, and a success toast.Mobile stacks into Chat | Player | Export tabs; keyboard shortcuts (Space, ←/→, L) work throughout.
Files built:
index.html,src/index.css,src/lib/types.ts,src/lib/format.ts,src/lib/mock.ts,src/lib/cuts.ts,src/state/cutter.tsx,src/components/Icons.tsx,src/components/ui.tsx,src/components/Landing.tsx,src/components/ChatPanel.tsx,src/components/PlayerStage.tsx,src/components/Timeline.tsx,src/components/ProjectPanel.tsx,src/components/Toast.tsx,src/components/TopBar.tsx,src/App.tsx
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.