CyberRealistic XL V8 Inpaint โ€” LocalMuse Core ML

This repository contains a Core ML build for on-device Inpaint in LocalMuse. It is derived from Cyberdelia's native nine-channel CyberRealistic XL v8.0-Inpainting checkpoint (Civitai model 312530, version 2462401).

Runtime profile

  • Workflow: dedicated SDXL Inpaint only
  • Resolution: fixed 1024ร—1024
  • UNet contract: [2, 9, 128, 128] โ†’ [2, 4, 128, 128]
  • UNet: 6-bit palettized, split Core ML program
  • Text encoders: checkpoint CLIP-L FP16 (non-finite final layer repaired from pinned SDXL Base) and checkpoint OpenCLIP-bigG 8-bit
  • VAE encoder/decoder: checkpoint-specific FP16
  • Recommended sampler: DPM++ 2M SDE Karras
  • Recommended settings: 30 steps, CFG 4, denoising strength 0.80
  • Device class: nominal 8 GB RAM or higher

Inference runs locally on the user's device after download. The source checkpoint is not included in this repository. Exact sources, revisions, hashes and converter details are recorded in PROVENANCE.json.

Source and attribution

License

The source version is identified by its author as CreativeML Open RAIL++-M with a version-specific Addendum. Both apply to this conversion:

Users are responsible for complying with the license, Addendum and applicable law.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support