LocalMuseAI/coreml-lustify-apex-inpaint-6bit

Lustify Apex Inpaint — Core ML 6-bit

Core ML conversion of LUSTIFY! APEX INPAINTING by coyotte for local, on-device inference in LocalMuse on iOS.

This repository contains compiled inference resources only. It does not contain the original single-file checkpoint. LocalMuse did not create a new model merge: the conversion started from the checkpoint already published by the original author and changed its runtime format and weight representation for Core ML.

Runtime contract

• Workflow: native SDXL Inpaint
• Resolution: 1024×1024
• UNet: dedicated 9-channel input, 4-channel output, split into two chunks
• Classifier-free guidance batch: 2
• UNet precision: mixed FP16 + 6-bit palettization
• TextEncoder: FP16
• TextEncoder2: mixed FP16 + 8-bit palettization
• VAE: source-checkpoint VAE, compiled with Float32 storage/compute
• Minimum deployment target: iOS 17
• Intended device class: nominal 8 GB RAM or higher

Recommended starting settings from the source model:

• Scheduler: DPM++ 2M SDE Karras
• Steps: 30
• CFG: 2.5–4.5 (3.5 default in LocalMuse)

Inference runs locally on the device after the model is downloaded.

Source and provenance

• Original author: coyotte
• Original model/version: Civitai model 573152, version 2875936
• Original filename: lustifyNSFWCheckpoint_apexINPAINTING.safetensors
• Original SHA-256: b6568b2da20a20a5706c3dde7aa3ddb43e8e2f570772b76080c2405caf9c91bf

Exact conversion metadata and hashes for every downloadable runtime file are recorded in PROVENANCE.json.

License

The source page identifies the license as CreativeML Open RAIL++-M with a source-specific Addendum. See LICENSE.md and SOURCE_ADDENDUM.md. Nothing in this repository grants rights beyond those source terms.

Content notice

This checkpoint can generate adult content. Use it only where lawful and appropriate, and keep LocalMuse filtering enabled unless you intentionally accept responsibility for disabling it.