mrbizarro /Phosphene

About Phosphene

Phosphene runs video, image and music models on your Mac, with no cloud or API keys. It's Apple Silicon only, built on MLX instead of PyTorch or CUDA. You get two video engines, stills, music and character LoRA training in one panel. Full character training needs 64 GB of memory; smaller Macs get cut-down recipes.

Best for: Mac users who want to render video with synced audio and train their own character LoRAs on their own Mac.

• Renders video with synced audio (lip-sync, footsteps, ambience) in one pass
• Switches between LTX-Video 2.5 and Hailuo H3 engines from the header
• Trains character LoRAs from 15–500 images, with local Gemma 3 12B captions
• Edits stills with Qwen-Image-Edit and up to three reference images
• Copies motion, camera moves and poses from a clip with Motion Control
• Queues renders and training over a local HTTP API on 127.0.0.1:8198

1. 1
   Install Pinokio, then install Phosphene from it. The base install fetches LTX-Video 2.5 (about 27.5 GB).
2. 2
   Add Hailuo H3 (about 75 GB) from the Pinokio sidebar if you want dialogue shots.
3. 3
   Install the YuE2 music engine (about 11 GB) from Audio → Compose.
4. 4
   Launch the panel and pick your engine with the segmented control in the header.
5. 5
   Queue a render over HTTP: the command below

   curl -s -X POST http://127.0.0.1:8198/queue/add --data-urlencode "mode=t2v" --data-urlencode "prompt=..."

   Copy