| --- |
| license: openrail++ |
| tags: |
| - text-to-image |
| - stable-diffusion-xl |
| - dreamshaper |
| - lightning |
| - coreml |
| - apple-neural-engine |
| - palettized |
| base_model: |
| - Lykon/dreamshaper-xl-lightning |
| pipeline_tag: text-to-image |
| library_name: ml-stable-diffusion |
| --- |
| |
| # DreamShaper XL Lightning - Core ML 6-bit |
|
|
| An iOS-oriented Core ML distribution of [Lykon/DreamShaper XL Lightning](https://huggingface.co/Lykon/dreamshaper-xl-lightning), prepared for on-device inference in the LocalMuse app. |
|
|
| No model training was performed by LocalMuseAI. The original model weights and authorship belong to Lykon and the upstream SDXL/SDXL-Lightning contributors. This repository only distributes the compiled Core ML representation and mobile-oriented weight palettization. |
|
|
| ## Runtime format |
|
|
| - Native resolution: 1024 x 1024 |
| - Apple Core ML compiled models (`.mlmodelc`) |
| - SPLIT_EINSUM attention |
| - Two 6-bit palettized UNet chunks, each below 1 GiB |
| - 6-bit palettized second text encoder |
| - VAE decoder for text-to-image |
| - VAE encoder for image-to-image and face-detail workflows |
| - CLIP vocabulary and merges |
| |
| LocalMuse uses a four-step DPM schedule, Neural Engine denoising, and GPU-preferred VAE execution. The package is intended for recent iPhones with at least 8 GiB of physical memory. |
| |
| ## License and attribution |
| |
| - Base model: [Lykon/dreamshaper-xl-lightning](https://huggingface.co/Lykon/dreamshaper-xl-lightning) |
| - Core ML tooling: [apple/ml-stable-diffusion](https://github.com/apple/ml-stable-diffusion) |
| - License: CreativeML Open RAIL++-M; see `LICENSE.md` |
| - This mirror imposes no additional restrictions. |
| |
