stable-audio-3-small-music-coreai-ios

Stable Audio 3 Small Music — Core AI, precompiled for iPhone

Powered by Stability AI

A trimmed, device-precompiled build of arraypress/stable-audio-3-small-music-coreai, which is a community conversion of stabilityai/stable-audio-3-small-music to Apple's Core AI format. The weights are unchanged from the source checkpoint. This build removes graphs that a phone app does not call and ships the result compiled ahead of time, so that first load on device takes seconds instead of minutes.

What's here

One .aimodelc bundle compiled for the h18p architecture (Apple A19 and A19 Pro, iPhone 17 family), holding four inference functions, plus the T5Gemma tokenizer:

The audio encoder graphs and the remaining decoder lengths from the source conversion were removed to reduce size. See NOTICE for the full modification lineage. The sampling loop (ping-pong sampler, 8 steps, log-SNR schedule) runs on the host; the source conversion's model card documents it fully.

Requirements

• iOS 27 or later, on an A19 or A19 Pro device
• Load with AIModel(contentsOf:) from the Core AI framework

License

Stability AI Community License (see LICENSE.md). The T5Gemma text encoder is subject to the Gemma Terms of Use (see LICENSE_GEMMA.md). All credit for the model to Stability AI, and for the Core AI conversion to arraypress.