Phosphene, Local Video and Audio Generation

Phosphene has been released as an open-source desktop panel for Apple Silicon Macs. It offers one-click installation via Pinokio.

It runs Lightricks' LTX 2.3 natively on top of MLX, generating video and audio together in a single forward pass. It is designed with frame-level timing coupling so that footsteps, dialogue, and ambient sound stay synchronized with the video.

Supported modes are as follows.

• Text → video
• Image → video
• First-frame / Last-frame
• Extend: appends to an existing clip while maintaining audio continuity

Prompts can be rewritten with a local Gemma 3 12B 4-bit text encoder, and it operates offline.

There are three quality tiers.

• Draft: half resolution, about 2 minutes
• Standard: 1280×704, about 7 minutes, Q4 distilled (25 GB)
• High: Q8 two-stage + TeaCache, about 12 minutes, an additional 25 GB, required for FFLF

The hardware is Apple Silicon only. It detects RAM at boot and limits features accordingly, with 32 GB supporting Compact and 64 GB supporting full features.