Box 3.2.0
New: On-device music & sound generation
The headline of this release — make music and sound effects from a text description,
completely on your device. No internet, no account, nothing leaves your phone. Just
describe what you want to hear and Box generates it, fully offline. You'll find it all
under the new Sound tab, and i must say it is mind boggling for this to be running on a phone.
- SoundGen — quick text-to-audio for short clips and sound effects. Type a prompt, get
audio in seconds. - SoundGen HD — higher-quality music and audio, up to ~24 seconds. Noticeably richer
detail for when quality matters. - SoundGen HD Long — long-form generation: full tracks up to ~3 minutes, for proper
tracks rather than clips. expect 5-10 mins for a full track.
Every tier lets you set the length, then play, save, or share the result right from the
app. Pick a vibe, hit Generate, and it's yours — 100% private and offline.
First use downloads the generator for that tier; after that it runs entirely on-device
with no connection.
New: Identify (on-device image recognition)
Point Box at a photo and it tells you what's in it — recognizing 1000+ everyday objects,
animals, and scenes. Pick from your gallery or snap a new shot. Runs fully offline,
with hardware acceleration on supported devices.
Refreshed interface
- Tabs reorganized for clarity: Sound (audio features) and Core (chat & assistant).
- Retro pixel-font accents on category headers and Settings.
- New Box cube app icon in the menu.
- A smoother "initializing" animation when a feature spins up.
Chat improvements
- Conversation memory on reopen — reopening a chat now replays recent context to the
model, so it remembers where you left off (new chats still start fresh). - Faster, more reliable responses on accelerated hardware, with a fix for an occasional
second-turn stall.
Fixes & polish
- Various performance,stability and memory improvements across several features.