Add Core ML inpainting model card

README.md

@@ -0,0 +1,41 @@
+# Clover Image Tiny Inpaint — Core ML
+
+This export is the iPhone 15 target for the 9-channel Clover Image Tiny
+inpainting U-Net. The conversion is pinned to the Apple Stable Diffusion
+converter vendored under `.coreml-converter` and targets iOS 18.
+
+## Resource contract
+
+The app-managed `Resources` directory contains:
+
+```text
+TextEncoder.mlmodelc
+VAEEncoder.mlmodelc
+VAEDecoder.mlmodelc
+SafetyChecker.mlmodelc
+UnetChunk1.mlmodelc
+UnetChunk2.mlmodelc
+vocab.json
+merges.txt
+```
+
+The U-Net input is `[1, 9, 64, 64]`, ordered as noisy latent, mask, and
+masked-image latent. The VAE encoder is required to create the final four
+channels on device. White mask pixels are regenerated and black pixels are
+preserved.
+
+## Convert and validate
+
+```bash
+coreml-tools/convert_inpaint.sh coreml-inpaint artifacts/clover-image-tiny-inpaint
+coreml-tools/smoke-test-inpaint.sh artifacts/clover-image-tiny-inpaint coreml-inpaint
+```
+
+The validator checks the 9-channel Diffusers configuration, required compiled
+resources, the fixed Core ML input shape, finite output, and random-input
+PyTorch/Core ML parity. The all-accelerators path is used because the macOS
+CPU-only Core ML runtime can return NaNs for this MLProgram even though the
+Neural Engine-capable path is valid.
+
+The resulting `Resources` directory can be published as the companion model
+repo [`neonforestmist/Clover-Image-Tiny-Inpaint-CoreML`](https://huggingface.co/neonforestmist/Clover-Image-Tiny-Inpaint-CoreML).