cardbud/dreamshaper-xl-lightning-coreml-ios

DreamShaper XL Lightning — Core ML for the iPhone Neural Engine

A Core ML conversion of Lykon/dreamshaper-xl-lightning (DreamShaper XL by Lykon, built on SDXL-Lightning by ByteDance and Stable Diffusion XL 1.0 by Stability AI), derived from the original model to run on the iPhone Neural Engine.

What was changed

These files are modified versions of the original model:

• Converted to Core ML with Apple's ml-stable-diffusion (torch2coreml, --attention-implementation SPLIT_EINSUM --chunk-unet --latent-h 96 --latent-w 96).
• Output size 768 × 768.
• The UNet is split into two chunks and its weights palettized to 6 bits (k-means). Text encoders are float16.
• The VAE decoder is the one from Apple's coreml-stable-diffusion-xl-base-ios, which uses madebyollin/sdxl-vae-fp16-fix so it runs in float16.

Files

compiled/ holds the compiled models in the layout Apple's StableDiffusionXLPipeline expects: TextEncoder.mlmodelc, TextEncoder2.mlmodelc, UnetChunk1.mlmodelc, UnetChunk2.mlmodelc, VAEDecoder.mlmodelc, vocab.json, merges.txt. Load with computeUnits = .cpuAndNeuralEngine.

Suggested settings: 4 steps, guidance scale 1–2.

License

Same license as the original model: CreativeML Open RAIL++-M (the SDXL 1.0 license). The use-based restrictions in Attachment A of the license apply to this model and to anything derived from it, and must be passed on to its users. The original model's terms and permissions, as published by its author, apply unchanged. As on the original model's page:

• Use without crediting the creator: allowed
• Share merges and derivatives: allowed, also under different permissions
• Sell images you generate: allowed
• Run on generation services: allowed
• Sell this model or merges: not allowed