NextGenUp 1.2.0 — Audio Studio, Video Tools, and security hardening

NextGenUp 1.2.0 — Audio Studio, Video Tools, and security hardening

NextGenUp now covers images, audio, and video. The sidebar is organized into three sections, every new tool runs locally, and the optional AI models download on first use so the installer stays the same size.

New: Audio section

• Vocal Remover — split any song into a vocals stem and an instrumental stem with MDX-Net (Kim Vocal 2). Karaoke tracks, acapellas, samples. Preview the original and both stems, save each, or download both as a zip. A free, offline alternative to lalal.ai and Moises.
• Noise Remover — remove background noise from speech recordings with DeepFilterNet3. This runs the official DeepFilterNet export and its output is numerically identical to the reference implementation. Podcasts, voice memos, interviews.
• Mastering — one-click automatic mastering that measures the track (loudness, true peak, dynamics, spectral balance, stereo width), decides every setting from what it hears, and explains each decision in plain language. Includes a before-vs-after table, spectral balance charts, a waveform, a loudness-matched A/B player, and a tuning panel (tone target, correction strength, bass, treble, de-hiss, loudness target, stereo width) seeded from the automatic decisions. Runs entirely in the app window; outputs 24-bit WAV.
• Tag Editor — view, strip, or replace MP3 and WAV metadata with byte-level surgery. The audio is copied unchanged and verified bit-for-bit identical.

New: Video Tools

• Remove Noise / Remove Music / Remove Voice — clean or separate a video's audio track and remux it with the untouched video stream. The picture is copied, never re-encoded.
• Compress — two-pass H.264 encode to a target size, with presets for Discord (10 MB), email (25 MB), and WhatsApp (100 MB).
• Make GIF — palette-optimized animated GIFs with fps and width options.
• Video Upscale 4K now supports bulk queues with a Basic/Pro mode selector.

Security

• File-writing endpoints (/api/save, /api/save-blob) and /api/open now require a per-session token sent as a custom header, which makes cross-origin requests fail the CORS preflight in every browser. Request origins are validated as a second layer, and the desktop backend binds to 127.0.0.1 only (the self-hosted web mode keeps optional LAN access). Thanks to @sdwwsrimo-web for the responsible report in #2.

Fixes

• Uploads with non-Latin filenames (for example Bengali) crashed on Windows because ffmpeg output was decoded with the system codepage. All subprocess output is now decoded as UTF-8.
• Every download now asks where to save: native dialogs in the desktop app, a Save As picker in Chromium browsers.

Models added (all MIT or Apache-2.0, downloaded on first use)

Kim Vocal 2 (~67 MB), DeepFilterNet3 (~8 MB). Mastering and the Tag Editor need no model.

Existing installations will be offered this update automatically on next launch.