LocalMuseAI/coreml-juggernaut-xl-v6-768-6bit

Juggernaut XL V6 — Core ML 768×768

This repository contains the fixed-shape Core ML runtime package used by LocalMuse for on-device text-to-image and image-to-image generation.

Runtime configuration

• Resolution: 768×768 (latent 96×96)
• Model: non-distilled Juggernaut XL V6
• UNet attention: SPLIT_EINSUM
• UNet storage: 6-bit palettized
• UNet execution target: CPU and Neural Engine
• VAE execution target in LocalMuse: GPU
• Classifier-free guidance: supported
• Recommended sampler: DPM++ 2M Karras
• Recommended steps: 30–40
• Recommended CFG: 3–7
• Minimum LocalMuse device memory policy: 8 GiB

The UNet is distributed as UnetChunk1.mlmodelc and UnetChunk2.mlmodelc. Both weight files are below 1 GiB. The package also contains both SDXL text encoders, the 768×768 VAE decoder, and the VAE encoder required for image-to-image and face-detail workflows.

Provenance

• Source checkpoint: RunDiffusion/Juggernaut-XL-v6
• Source revision: 95b07c32cf8a3b8fe958799cc121fffcf4362680
• Source checkpoint SHA-256: 1fe6c7ec54c786040cdabc7b4e89720069d97096922e20d01f13e7764412b47f
• 768×768 Core ML component reference: coreml-community/coreml-juggernautXL-v6_SDXL_8-bit
• Component reference revision: 98146a1e3f3ca26f07d780e91063a12ea130547e
• Apple conversion tools revision: e12202c1f6405b83918b58a5d097cd61e3e1f702
• Minimum Core ML deployment target: iOS 17

The model-specific UNet was converted from the pinned source checkpoint, palettized to 6 bits, split into two ML Programs, and compiled with Xcode. The fixed 768×768 SDXL text encoders and VAE components were taken from the pinned Core ML component reference. Every runtime file is distributed through a commit-pinned URL and authenticated by SHA-256 in the LocalMuse catalog.

License

Juggernaut XL V6 and this converted derivative are distributed under the CreativeML Open RAIL++-M terms in LICENSE.md. Review the source model card and license restrictions before using or redistributing the model.