MisterSnowCrash/sdxl-coreml-t1024

Core ML SDXL: DreamShaper XL v2 Turbo (1024, split_einsum, 6-bit)

Core ML conversion for on-device text-to-image on Apple Neural Engine (iOS / macOS), at 1024x1024 (SDXL's native resolution). Experimental; may be renamed or removed.

• Base model: Lykon/dreamshaper-xl-v2-turbo
• Converted with apple/ml-stable-diffusion torch2coreml
• 1024x1024 native, SPLIT_EINSUM attention, 6-bit palettization, UNet in two chunks (< 1 GB each), fp16-fix VAE
• Recommended: 6 steps, guidance 2.0, DPM-Solver++ (Karras)

License

CreativeML Open RAIL++-M, inherited from the base model (see the license text). The use restrictions in Attachment A apply to anyone using these weights or their outputs: no use that violates laws; no exploitation or harm of minors; no generation of false information to harm others; no personal identifiable information to harm; no defamation or harassment; no fully automated decisions that adversely affect legal rights; no discrimination; no exploitation of vulnerabilities of specific groups; no medical advice or results interpretation; no generation or dissemination of information for administration of justice or law enforcement purposes.

All credit to the original authors. This is a format conversion only; weights are unchanged apart from 6-bit palettization.