Upload sd-turbo-coreml (CommuniDev LLC)

LICENSE.md

@@ -4,7 +4,7 @@ This repository is a Core ML conversion of [stabilityai/sd-turbo](https://huggin
 made by CommuniDev LLC (format conversion and 6-bit palettization only; no retraining), plus an unmodified
 safety checker. See `NOTICE` for exactly what was changed.
 
-1. **Stability AI Community License** — covers `TextEncoder`, `Unet` and `VAEDecoder` (the SD-Turbo model).
+1. **Stability AI Community License** — covers `TextEncoder`, `Unet`, `VAEDecoder` and `VAEEncoder` (the SD-Turbo model).
    Full text, word for word: [`LICENSE-stability-ai-community.md`](LICENSE-stability-ai-community.md).
    Use must follow Stability AI's Acceptable Use Policy: https://stability.ai/use-policy.
    Commercial use requires registering with Stability AI (https://stability.ai/community-license) and is

NOTICE

@@ -3,13 +3,15 @@ This Stability AI Model is licensed under the Stability AI Community License, Co
 Powered by Stability AI.
 
 Attributions that apply to the Stability AI Materials: the notice above applies to TextEncoder.mlmodelc,
-Unet.mlmodelc and VAEDecoder.mlmodelc, which are derived from stabilityai/sd-turbo (revision b261bac6).
+Unet.mlmodelc, VAEDecoder.mlmodelc and VAEEncoder.mlmodelc, which are derived from stabilityai/sd-turbo
+(revision b261bac6).
 
 Changes made by CommuniDev LLC (2026-10-10): the Stability AI Materials were converted from their fp16
 safetensors to Core ML with Apple's ml-stable-diffusion converter (torch2coreml, commit 5a170d2),
 attention SPLIT_EINSUM_V2, 512 x 512; the UNet and text encoder weights were palettized to 6 bits with
-k-means (the text encoder's 77 x 77 attention-mask constant left uncompressed); the VAE decoder is fp16;
-everything was compiled to .mlmodelc. No retraining or fine-tuning. The VAE encoder is not included.
+k-means (the text encoder's 77 x 77 attention-mask constant left uncompressed); the VAE decoder and
+VAE encoder are fp16; everything was compiled to .mlmodelc. No retraining or fine-tuning. The VAE encoder
+was added to this repository on 2026-10-19, from the same 2026-10-10 conversion run.
 
 Other files, not Stability AI Materials:
 - SafetyChecker.mlmodelc: CompVis stable-diffusion-safety-checker, in Apple's Core ML conversion from

README.md

@@ -8,6 +8,7 @@ tags:
   - core-ml
   - sd-turbo
   - text-to-image
+  - image-to-image
   - ios
 ---
 
@@ -18,8 +19,9 @@ created by **Stability AI**: a fast, distilled version of Stable Diffusion 2.1 t
 in 1–4 steps.
 
 CommuniDev LLC did **not** train or fine-tune this model. We only changed its file format so it can run on
-an iPhone. It is published so that our iOS app [CommuniDream](https://communidev.com) can download it on
-first use and turn dream journal entries into pictures, entirely on the device.
+an iPhone. It is published so that our iOS apps can download it on first use and run it entirely on the
+device: [CommuniDream](https://communidev.com) turns dream journal entries into pictures, and
+[CommuniMuse](https://communidev.com) restyles the user's own photos (image-to-image).
 
 ## What we changed
 
@@ -29,11 +31,12 @@ Converted with Apple's [ml-stable-diffusion](https://github.com/apple/ml-stable-
 - Core ML, attention `SPLIT_EINSUM_V2` (Neural Engine layout), 512 × 512 images.
 - UNet and text encoder: **6-bit palettized** weights (k-means). The text encoder's 77 × 77 causal
   attention mask holds `-inf` and is left uncompressed.
-- VAE decoder: fp16. The VAE encoder is not included (text-to-image only).
+- VAE decoder and VAE encoder: fp16. The encoder (added 2026-10-19, from the same conversion run) turns a
+  starting picture into latents, for image-to-image.
 - Safety checker: the CompVis Stable Diffusion safety checker in Apple's Core ML conversion, unmodified.
 
-Modified files: `TextEncoder.mlmodelc`, `Unet.mlmodelc` and `VAEDecoder.mlmodelc` are converted, modified
-forms of the original checkpoint. Details are in `NOTICE`.
+Modified files: `TextEncoder.mlmodelc`, `Unet.mlmodelc`, `VAEDecoder.mlmodelc` and `VAEEncoder.mlmodelc`
+are converted, modified forms of the original checkpoint. Details are in `NOTICE`.
 
 ## Files
 
@@ -42,6 +45,7 @@ forms of the original checkpoint. Details are in `NOTICE`.
 | `compiled/Unet.mlmodelc` | 623 MB |
 | `compiled/TextEncoder.mlmodelc` | 304 MB |
 | `compiled/VAEDecoder.mlmodelc` | 95 MB |
+| `compiled/VAEEncoder.mlmodelc` | 65 MB |
 | `compiled/SafetyChecker.mlmodelc` | 580 MB |
 | `compiled/vocab.json`, `compiled/merges.txt` | 1.3 MB |
 
@@ -51,7 +55,9 @@ forms of the original checkpoint. Details are in `NOTICE`.
 
 With Apple's Swift `StableDiffusionPipeline`: DPM-Solver++ (linspace spacing, which starts at t = 999),
 2–4 steps, guidance 1.0–2.0, seed-reproducible. At guidance 1.0 the negative prompt has no effect.
-**The safety checker must stay on** (`disableSafety: false`); CommuniDream refuses to run without it.
+For image-to-image, pass a 512 × 512 starting image and a strength; the steps actually run are
+steps × strength, so keep that at 1 or more.
+**The safety checker must stay on**: CommuniDream and CommuniMuse refuse to run without it.
 
 ## Limitations
 
@@ -62,7 +68,7 @@ With Apple's Swift `StableDiffusionPipeline`: DPM-Solver++ (linspace spacing, wh
 
 ## Licence
 
-`TextEncoder`, `Unet` and `VAEDecoder` are under the **Stability AI Community License**
+`TextEncoder`, `Unet`, `VAEDecoder` and `VAEEncoder` are under the **Stability AI Community License**
 (`LICENSE-stability-ai-community.md`, word for word): use must follow Stability AI's
 [Acceptable Use Policy](https://stability.ai/use-policy), and commercial use requires registration with
 Stability AI. `SafetyChecker` is under **CreativeML Open RAIL-M** (`LICENSE-creativeml-openrail-m.txt`).

SHA256SUMS

@@ -18,5 +18,10 @@ ec4f6b87000f1f19c1237776aa4111d6c7e40832db3eee89eb2fa98fb3f686b5  compiled/VAEDe
 6b63ca07bbbe2cc8c3533d6a07ba1bdf4b5d350111a435e88856f0d1759a72bf  compiled/VAEDecoder.mlmodelc/metadata.json
 e0447f4f41d531f94ce20e81d204f0aace02a79aad8d4c0fb3b982a0822e00a6  compiled/VAEDecoder.mlmodelc/model.mil
 7cc4b185bde09496bc25e236de3a0da8d78ef83fc8a7e8fbe4435f3027413c6f  compiled/VAEDecoder.mlmodelc/weights/weight.bin
+7f2c36edf7102be7e09080b79baa9fadce60af00404c2efc0966deaa264a7067  compiled/VAEEncoder.mlmodelc/analytics/coremldata.bin
+43cd99c3612fd7b9df97eb9deea0c35050d66a488df8b4d56906685ab6649e95  compiled/VAEEncoder.mlmodelc/coremldata.bin
+ea02a31f66ba79604813cda56622bb2cd0605432b445e02e0b866909c0fac022  compiled/VAEEncoder.mlmodelc/metadata.json
+69a39088eceee9476f40145f0aff29e8d0ee9d6d017123f75f661551fac8957d  compiled/VAEEncoder.mlmodelc/model.mil
+993f3790072d02f4ba7baaa418b5135be56f99f280f05dfc33515be66dfb3b7d  compiled/VAEEncoder.mlmodelc/weights/weight.bin
 f526393189112391ce6f9795d4695f704121ce452c3aad1f5335cc41337eba85  compiled/merges.txt
 5047b556ce86ccaf6aa22b3ffccfc52d391ea4accdab9c2f2407da5b742d4363  compiled/vocab.json

compiled/VAEEncoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f2c36edf7102be7e09080b79baa9fadce60af00404c2efc0966deaa264a7067
+size 243

compiled/VAEEncoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43cd99c3612fd7b9df97eb9deea0c35050d66a488df8b4d56906685ab6649e95
+size 782

compiled/VAEEncoder.mlmodelc/metadata.json

@@ -0,0 +1,80 @@
+[
+  {
+    "shortDescription" : "Stable Diffusion generates images conditioned on text and\/or other images as input through the diffusion process. Please refer to https:\/\/arxiv.org\/abs\/2112.10752 for details.",
+    "metadataOutputVersion" : "3.0",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 8 × 64 × 64)",
+        "shortDescription" : "The latent embeddings from the unet model from the input image.",
+        "shape" : "[1, 8, 64, 64]",
+        "name" : "latent",
+        "type" : "MultiArray"
+      }
+    ],
+    "version" : "sd-turbo",
+    "modelParameters" : [
+
+    ],
+    "author" : "Please refer to the Model Card available at huggingface.co\/sd-turbo",
+    "specificationVersion" : 7,
+    "storagePrecision" : "Float16",
+    "license" : "OpenRAIL (https:\/\/huggingface.co\/spaces\/CompVis\/stable-diffusion-license)",
+    "mlProgramOperationTypeHistogram" : {
+      "Pad" : 3,
+      "Ios16.cast" : 1,
+      "Ios16.mul" : 2,
+      "Ios16.sub" : 22,
+      "Ios16.sqrt" : 22,
+      "Transpose" : 6,
+      "Ios16.square" : 22,
+      "Ios16.add" : 34,
+      "Ios16.reduceMean" : 44,
+      "Ios16.realDiv" : 22,
+      "Ios16.conv" : 28,
+      "Ios16.linear" : 4,
+      "Ios16.matmul" : 2,
+      "Ios16.batchNorm" : 21,
+      "Ios16.softmax" : 1,
+      "Ios16.reshape" : 49,
+      "Ios16.silu" : 21
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "stateSchema" : [
+
+    ],
+    "isUpdatable" : "0",
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 3 × 512 × 512)",
+        "shortDescription" : "The input image to base the initial latents on normalized to range [-1, 1]",
+        "shape" : "[1, 3, 512, 512]",
+        "name" : "x",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.version" : "7.2",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.source" : "torch==2.2.2"
+    },
+    "generatedClassName" : "Stable_Diffusion_version_sd_turbo_vae_encoder",
+    "method" : "predict"
+  }
+]

compiled/VAEEncoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:993f3790072d02f4ba7baaa418b5135be56f99f280f05dfc33515be66dfb3b7d
+size 68338112