README.md

---
license: openrail
library_name: coreml
pipeline_tag: text-to-image
base_model: ByteDance/Hyper-SD
base_model_relation: quantized
tags:
  - coreml
  - core-ml
  - ios
  - macos
  - apple
  - on-device
  - text-to-image
  - stable-diffusion
  - one-step
  - lora
  - arxiv:2404.13686
---

# Hyper-SD (1-Step) — Core ML

*ByteDance, 2024*

Single-step text-to-image from SD1.5 via TCD distillation. 512×512. Chunked UNet (6-bit).

<p><img src="https://huggingface.co/mlboydaisuke/Hyper-SD-1step-CoreML/resolve/main/media/1c6281beb3.png" alt="Hyper-SD (1-Step) demo" width="49%"> <img src="https://huggingface.co/mlboydaisuke/Hyper-SD-1step-CoreML/resolve/main/media/008e4341e0.png" alt="Hyper-SD (1-Step) demo" width="49%"></p>

Core ML conversion of [ByteDance/Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD) for on-device inference on iPhone, iPad and Mac. Converted with `coremltools`; the packages are stateless, so all sequencing and buffering lives in your Swift code.

| | |
|---|---|
| Task | text to image |
| Upstream | [ByteDance/Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD) |
| Packages | 4 |
| Download size | 905 MB |
| Minimum iOS | 17.0 |
| Peak RAM | ~1000 MB |

## Files

| File | Size | Compute units | SHA-256 |
|---|---:|---|---|
| `HyperSDTextEncoder.mlpackage.zip` | 216 MB | `cpuAndNeuralEngine` | `201b0fcc3573811a…` |
| `HyperSDUnetChunk1.mlpackage.zip` | 310 MB | `cpuAndNeuralEngine` | `279da11b8231aeeb…` |
| `HyperSDUnetChunk2.mlpackage.zip` | 290 MB | `cpuAndNeuralEngine` | `0a700d11a105da58…` |
| `HyperSDVAEDecoder.mlpackage.zip` | 87 MB | `cpuAndGPU` | `1260371542d845a2…` |
| `vocab.json` | 1 MB | `-` | `e089ad92ba36837a…` |
| `merges.txt` | 512 KB | `-` | `9fd691f7c8039210…` |
| **Total** | **905 MB** | | |

`compute_units` is not a suggestion -- it is the configuration the conversion was verified against. Moving a package to a different compute unit can silently change the numerics (FP16 attention overflow) or crash on the GPU.

## Download

```bash
hf download mlboydaisuke/coreml-zoo --include "hypersd/*" --local-dir ./hypersd
unzip './hypersd/hypersd/*.zip' -d ./hypersd
```

## Use in Swift

```swift
import CoreML

let config = MLModelConfiguration()
config.computeUnits = .cpuAndNeuralEngine   // as converted — see the table above

// Unzip the .mlpackage, drop it into your Xcode target and Xcode compiles it
// at build time:
let model = try HyperSDTextEncoder(configuration: config)

// ...or compile a downloaded .mlpackage at runtime:
let compiled = try await MLModel.compileModel(at: mlpackageURL)
let model = try MLModel(contentsOf: compiled, configuration: config)
```

> This model is split into 4 Core ML packages that are driven in sequence from Swift. Load them one at a time, copy the outputs out of the `MLMultiArray` buffers and release each model before loading the next — two large Core ML models resident at once will OOM on an iPhone.

## Demo

- **Sample app** — [`sample_apps/HyperSDDemo`](https://github.com/john-rocky/CoreML-Models/tree/master/sample_apps/HyperSDDemo), a standalone SwiftUI project.
- **Models Zoo** — this model is downloadable and runnable inside the [Models Zoo app](https://apps.apple.com/app/id6762083207) on the App Store, no build required.

## Conversion

- Script: [`convert_hypersd.py`](https://github.com/john-rocky/CoreML-Models/blob/master/conversion_scripts/convert_hypersd.py)
- Pitfalls hit during conversion (FP16 overflow, ANE buffer limits, stride handling): [`docs/coreml_conversion_notes.md`](https://github.com/john-rocky/CoreML-Models/blob/master/docs/coreml_conversion_notes.md)
- Model index: [CoreML-Models](https://github.com/john-rocky/CoreML-Models)

## License

The conversion inherits the upstream license: **OpenRAIL-M**.

## Credits

- Upstream authors: [ByteDance/Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD), 2024
- Core ML conversion: john-rocky (Daisuke Majima)