Features

Everything in the beta.

CueTwin is built with Electron, React and Remotion, and uses FFmpeg for media work. Here is what the beta build does, in the app’s own terms.

A one-minute tour of the real CueTwin 1.2.0 app: import, timeline, titles, a split, a transition, a punch-in, cards and export. Captions and zooms were added afterwards, and the voice picker is blurred; nothing else in the interface was changed.

Assistant

Edit by typing, speaking or scripting

Ask for an edit in plain words, or dictate it with your microphone. The assistant plans the change and CueTwin applies it to your timeline, where you can review it, adjust it or undo it.

  • CueTwin (built-in): a local llama.cpp server that CueTwin starts and manages itself. It uses Qwen3.5 9B (“Pro”) when your GPU has room and Qwen3.5 4B (“Lite”) otherwise. No API key, no Ollama, no usage credits.
  • Script-driven narration: have the assistant polish a narration script, review the draft, then approve it to generate speech in your chosen voice and place the take on an audio track.
  • Optional backends: Claude Code CLI, Claude with an API key, Cursor CLI, local Ollama, or any OpenAI-compatible service (OpenAI, OpenRouter, Groq, custom). These need your own account or installation; cloud services receive your prompts and are billed by their providers.

The built-in assistant uses CueTwin’s editing tools and instructions with local Qwen base models, downloaded on first launch (roughly 8 GB). It is not a custom-trained model.

Editing

Multi-track timeline

A familiar layout: media library, Remotion-powered preview, multi-track timeline and the AI side panel.

  • Import video, image and audio; drag onto the timeline; move, resize, split and delete clips.
  • Text overlays and per-clip speed, shown live in the preview.
  • Undo and redo. Removing media from a project never touches the file on disk.
  • Projects are folders: recordings, voiceover, media and exports live together, so a project moves and backs up as one directory.
  • New audio lands at the playhead or where you drop it. Recordings, generated speech and imported audio use a free audio track for that range, and CueTwin adds a track when every track is busy. Undo removes it again.

Transitions:

  • Dissolve
  • Fade
  • Morph
  • Box
  • Wipe
  • Slide
  • Zoom
  • Fast Zoom

Projects

Move a whole project with CuePack

Project → Export Project (.CuePack) packs the current edit into one file, including changes you haven’t saved yet. Copy it to another PC and use Import Project (.CuePack) to open it there.

  • Includes every video, audio file and image in the media bin (used or not), all timelines and sections, media-bin folders, narration scripts and generated narration audio.
  • Import creates a new project folder, verifies the files, relinks the media on that computer and opens the project in a tab. The original project is never changed.
  • Media is stored without recompression and streamed to disk, so large projects don’t fill your memory. CueTwin checks free disk space first and rejects damaged or incomplete bundles before opening them.
  • Each export or import shows its own progress, with a cancel button, in the top-right tray. You can keep editing, or start another export, while one runs.

A CuePack holds project assets only: not your settings, API keys, trained voice models, installed fonts or the app itself.

Cleanup

Trim silence and dead air

Trim silence on the timeline cuts dead air, leftover hiss and clicks from the selected clips, then closes the gap.

For long recordings, ask the assistant to “cut the useless parts out of this recording”. It runs FFmpeg silence detection and rebuilds the timeline from the parts worth keeping. Thresholds, minimum silence, padding and merge gaps can all be steered.

Screen recordings

Smart edit

Silence detection can’t help with a narration-free screen recording of an AI agent driving another app. Smart edit scores each half of the screen separately:

  • App viewport changing: kept at normal speed.
  • Only the agent’s chat pane changing: sped up (4x by default).
  • Nothing changing: cut.
  • An unrelated app fullscreen, or a different project in the chat pane: cut.

Speed shows as a badge on each clip and plays live in the preview, so you can judge a ramp by scrubbing.

Canvas

Aspect ratios and resolutions

Aspect ratios:

  • 16:9
  • 9:16
  • 1:1
  • 4:5
  • 21:9

Resolutions:

  • 720p
  • 1080p
  • 1440p
  • 4K

Capture

Screen and game recording

  • Display records a whole monitor (also the right choice for a game in exclusive fullscreen); Window / Game records one window.
  • Up to 4K at 24–120 fps, three quality tiers, system audio and a microphone meter.
  • Saved as MP4 through the hardware encoder (NVENC, QSV or AMF), falling back to libx264. Recordings go into the project’s folder and onto the timeline.
  • A recording survives a crash: chunks are written to disk as they arrive, and the next launch recovers the footage.

Export

FFmpeg export

Exports render for real with FFmpeg, cut from your original sources rather than preview proxies, with per-clip speed applied. A hardware encoder (NVENC, QSV or AMF) is used when present, falling back to libx264.

Formats, as supported by the included FFmpeg build:

  • MP4 - H.264
  • MP4 - H.265 / HEVC
  • MOV - H.264
  • MOV - H.265 / HEVC
  • MOV - Apple ProRes 422 HQ
  • WebM - VP9

Frame rates 24, 25, 30, 50 and 60 fps. Audio as AAC, Opus, or uncompressed WAV/PCM (16- or 24-bit).

Free beta: exports up to 1080p (short edge up to 1080 pixels, long edge up to 2560, portrait included). Higher resolutions will need a paid Creator licence once licences go on sale.

Voice

Local voice cloning

In Voice Studio, record samples with your microphone or add existing audio and video files, then choose:

  • Instant clone: picks a clean reference from your samples. No long training recording needed.
  • Professional clone: trains a local voice from about ten minutes of usable speech. CueTwin compares candidates and saves a voice only if it passes automated checks for identity, wording, pacing and noise.

A trained voice gets a Voice version switch: Trained voice uses what was learned, and Original reference clone keeps the single-clip version for comparison.

Free beta: instant cloning is included. Professional voice training will need a paid Creator licence once licences go on sale.

Speech is generated locally with Qwen3-TTS, and a local Whisper model reads each take back to catch wrong words. Recordings and training stay on your PC. Only clone voices you have permission to use, and listen to generated speech before you publish it.

Languages (Qwen3 voices):

  • English
  • Chinese
  • German
  • Italian
  • Portuguese
  • Spanish
  • Japanese
  • Korean
  • French
  • Russian

An optional multilingual model (separate one-time download) adds instant cloning in: Hindi, Arabic, Danish, Dutch, Finnish, Greek, Hebrew, Malay, Norwegian, Polish, Swahili, Swedish, Turkish.

Privacy

Local once it’s set up

Editing, the built-in assistant and voice generation run on your own computer once the app and its models are installed. You need the internet for the first-launch AI model download (roughly 8 GB), for the first voice setup (Python packages and speech model downloads), for optional cloud AI backends, and for licence checks once paid licences arrive.

Not in the beta: AI background music is a development-only feature and isn’t part of the public beta.