jc-builds/sd2-inpainting-coreml

Stable Diffusion 2.1 Inpainting (Core ML)

This is a Core ML conversion of Stable Diffusion 2.1 Inpainting optimized for on-device inference on Apple Silicon.

Model Details

Property	Value
Original Model	Stable Diffusion 2.1 Inpainting
Parameters	865 million
Resolution	512x512
Format	Core ML (.mlmodelc)
Compute Units	CPU + Neural Engine (Split Einsum)

Intended Use

This model is optimized for:

• iOS/macOS Deployment: Native Apple Silicon inference
• Inpainting Tasks: Fill in masked areas of images
• Object Removal: Remove unwanted objects from photos
• Object Replacement: Replace specific parts of images

Capabilities

• Mask-based Editing: Paint areas to regenerate
• Seamless Blending: Natural integration with original content
• Prompt-guided Generation: Control what appears in masked areas
• High Quality Output: SD 2.1 quality improvements

Bundle Contents

• TextEncoder.mlmodelc - CLIP text encoder
• Unet.mlmodelc - 9-channel UNet (latent + masked image + mask)
• VAEDecoder.mlmodelc - Variational autoencoder decoder
• VAEEncoder.mlmodelc - Variational autoencoder encoder
• vocab.json - Tokenizer vocabulary
• merges.txt - BPE merges

Usage

Compatible with Apple's ml-stable-diffusion framework and iOS apps using Core ML.

License

This model is released under the CreativeML Open RAIL++-M license.

Attribution

• Original Model: Stable Diffusion 2.1 Inpainting by Stability AI
• Core ML Conversion: jc-builds