VideoMagic
WINDOWS FIRST · LOCAL BY DEFAULT

Turn a clip and script into a finished AI voiceover video.

Import a local video, write the narration, place each sentence on a visual timeline, preview the voice, then render subtitles, ducking and the final MP4 on your own PC.

Download for Windows Latest release View source on GitHub
✓ Source video stays local ✓ CUDA or CPU ✓ No cloud TTS API required
VideoMagic Local AI ready
▶
00:07.2 / 00:12.0
NARRATION SCRIPT

不是哥们，这猴子是真的没拿自己当外人。

上来先把游客的可乐抢了，下一秒更离谱……

NARRATION Waveform · manual timing
01这猴子是真的没拿自己当外人
02下一秒更离谱
Local mediaVideo never needs to be uploaded for the core workflow.
Real timelineDragging a block changes where narration is physically rendered.
Native outputH.264/AAC MP4, synchronized subtitles and smart ducking.
THE WORKFLOW

From source clip to finished commentary, in one workspace.

The simple path stays simple. Use automatic timing when it is good enough, or open the creator timeline when you want control.

01

Import

Choose or drag in MP4, MOV, MKV, WEBM or M4V. VideoMagic probes duration and audio locally.

02

Write

Paste Mandarin narration and choose a local voice preset. Preview the voice before committing to a render.

03

Time it

Auto-spread sentences, or drag and resize narration blocks against the source-audio waveform.

04

Render

Generate voice, synchronized subtitles, source-audio ducking and the final MP4 with progress and cancel controls.

CREATOR TOOLS

Enough control to edit the story, without becoming a full NLE.

01

Visual narration timeline

Sentence-level blocks are real render inputs. Drag a block to move narration, pull either edge to change its timing window, or type exact start and end values.

01 · Intro 02 · Punchline
02

Source waveform

FFmpeg extracts a lightweight local waveform so narration can be placed around source-audio beats and quiet gaps.

03

Smart ducking

Source audio drops only while narration is speaking, using sidechain compression instead of lowering the whole clip.

04

Synchronized subtitles

Subtitle timing comes from the exact same narration timeline, then burns into the final H.264 render.

05

Persistent projects

Save native .vmagic projects with source, script, voice settings and manual timeline positions.

06

Warm local AI

The local worker stays alive between previews and renders so repeated voice generation does not reload the model every time.

LOCAL-FIRST ARCHITECTURE

The heavy work happens on your machine.

VideoMagic separates the desktop UI from the local AI/media engine so voices and render capabilities can evolve without rewriting the product.

DESKTOPReact + TypeScriptCreator workspace
→
NATIVE BRIDGETauri 2 + RustFiles, jobs, permissions
→
LOCAL ENGINEPython 3.11Persistent worker
→
KokoroMandarin TTS
FFmpegWaveform · mix · subtitles · export
Your source fileremains on the PC during the core voiceover workflow.
Large AI datacan live on a drive you choose instead of filling the system disk.
GPU accelerationprefers NVIDIA CUDA/NVENC when available and can fall back to CPU.
DOWNLOAD

VideoMagic for Windows

Small desktop installer. On first run, choose where the local Python runtime, AI model cache and FFmpeg should live.

Windows 10/11 x64 NVIDIA GPU recommended CPU fallback available
Download installer Latest release Release notes & checksums ↗

Current public builds are unsigned. Windows SmartScreen may show a warning.

CURRENT STATUS

Windows-first V0.1, actively evolving.

Working now

Local Mandarin TTS, auto/manual timing, timeline waveform, resize handles, subtitles, ducking, CUDA/CPU selection, render history and project files.

Next

Per-segment audio caching, stronger/custom voices, clean-machine validation, code signing and updater hardening.

Not pretending yet

This is not a full Premiere replacement and it is not a cloud collaboration suite. The focus is fast local AI commentary creation.