Tiny Watch Image Generator LCM256 Core ML Artifacts

This repository contains external Core ML artifacts for the WatchPipelineSmokeApp LCM256 baseline in lube8163-lab/tiny-watch-image-generator.

The GitHub repository intentionally keeps large .mlmodelc and .mlpackage artifacts out of Git. Download these files when you want to build or evaluate the LCM256 Apple Watch pipeline locally.

Files

watch_lcm256_v0.1.0/watchos_app_assets.zip
- Drop-in compiled .mlmodelc resources for WatchPipelineSmokeApp.
- Includes the 16-part LCM UNet, 256px 4-bit decoder, CLIP text encoder, tokenizer files, and LCM256Assets/lcm_scheduler.json.
watch_lcm256_v0.1.0/coreml_packages.zip
- Source .mlpackage artifacts used by the Mac-side quality evaluator and future rebuilds.
watch_lcm256_v0.1.0/manifest.json
- Version, source commit, file list, and install targets.
watch_lcm256_v0.1.0/sha256sums.txt
- Checksums for release files.

Quick Start

git clone https://github.com/lube8163-lab/tiny-watch-image-generator.git
cd tiny-watch-image-generator
python3 tools/fetch_watch_lcm256_assets.py
open watchos_example/TinyImageWatchApp.xcodeproj

Then select the WatchPipelineSmokeApp scheme in Xcode.

Baseline

Pipeline: LCM256 6b
Resolution: 256x256
Steps: 4
Guidance: 6
Watch compute path: CPU-only Core ML
Text encoder: transient CLIP text encoder
Denoiser: 16 streamed 6-bit UNet chunks
Decoder: 256px 4-bit VAE decoder

Provenance

The converted artifacts are derived from SimianLuo/LCM_Dreamshaper_v7, which is an LCM distilled from Lykon/dreamshaper-7.

Please review the upstream model cards and licenses before using these artifacts in your own project. This upload is intended for research and technical demo use.

Downloads last month: -; Downloads are not tracked for this model. How to track