diff --git a/CHANGELOG.md b/CHANGELOG.md
index 8a53666..0d168e9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,79 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.11.0] - 2025-10-12
+
+# MFLUX v.0.11.0 Release Notes
+
+### 🎨 New Model Support
+
+- **Qwen Image Support**: Added support for the Qwen Image text-to-image model, enabling a new generation of visual content creation
+- **New command**: `mflux-generate` now supports Qwen models for image generation
+- **Qwen-specific features**: Full LoRA support with Qwen naming conventions, img2img support, and optimized weight handling
+- **Qwen-Image-mflux-6bit Model**: Added [filipstrand/Qwen-Image-mflux-6bit](https://huggingface.co/filipstrand/Qwen-Image-mflux-6bit) quantized model to HF
+
+### 🏗️ Major Architecture Improvements
+
+- **Package Restructure**: Complete reorganization of the codebase to support multiple model architectures
+  - Moved from flat structure to organized `models/` hierarchy (`models/flux/`, `models/qwen/`, `models/depth_pro/`)
+  - Better separation of concerns with dedicated model, variant, tokenizer, and weight handler modules
+  - Improved maintainability and extensibility for future model additions
+- **Namespace Package**: Converted mflux to a namespace package (in preparation for mflux.mcp extension)
+- **Common Module**: Extracted shared functionality into `models/common/` for better code reuse
+  - Unified LoRA handling across different model types
+  - Shared attention utilities
+  - Common download and weight management utilities
+
+### 📊 Metadata Enhancements
+
+- **XMP/IPTC Metadata Support**: Added comprehensive metadata support for professional workflows
+  - Write XMP and IPTC metadata to generated images
+  - Industry-standard metadata formats for better compatibility with professional image tools
+  - Enhanced metadata reading and writing capabilities
+- **New `mflux-info` command**: Display detailed metadata information from generated images
+  - View generation parameters, model information, and settings
+  - Extract metadata from any mflux-generated image
+  - Professional-grade metadata inspection
+
+### 🔧 Scheduler System
+
+- **Scheduler Interface**: Introduced a new scheduler abstraction for better extensibility
+  - Clean interface for implementing custom sampling schedulers
+  - Foundation for future scheduler additions (Euler, DPM++, etc.)
+  - Current implementation: Linear scheduler (existing behavior preserved)
+- **Scheduler Selection**: Added `--scheduler` command-line argument for choosing schedulers
+
+### 🐛 Bug Fixes
+
+- **Non-Quantized Model Loading**: Fixed critical bug where locally saved non-quantized models failed to load properly
+- **Model Weight Handling**: Improved weight loading reliability for edge cases
+
+### 🔧 Developer Experience
+
+- **MLX 0.29.2 Support**: Updated MLX dependency to support the latest version (mlx>=0.27.0,<0.30.0)
+- **Python 3.13 Support**: Unblocked sentencepiece and torch dependencies for Python 3.13
+  - Updated dependency specifications for better Python 3.13 compatibility
+  - Ensured smooth experience on latest Python versions
+- **Test Improvements**: Enhanced image comparison logic to allow similar images that are "close enough"
+  - More robust test suite that accommodates minor numerical differences
+  - Reduced false positives in image generation tests
+- **CI Updates**: Removed Claude CI agent (replacement coming soon)
+
+### 🔄 Breaking Changes
+
+⚠️ **Import Path Changes**: Due to the package restructure, some internal import paths have changed. If you're using mflux as a library and importing internal modules directly, you may need to update your imports:
+- Flux modules moved from `mflux.flux.*` to `mflux.models.flux.*`
+- Common utilities moved to `mflux.models.common.*`
+- CLI tools remain unchanged and fully backward compatible
+
+### 👩‍💻 Contributors
+
+- **Filip Strand (@filipstrand)**: Qwen model support, package restructure, core development
+- **Alessandro Rizzo (@azrahello)**: XMP/IPTC metadata support, info command implementation
+- **Anthony Wu (@anthonywu)**: Scheduler interface, namespace package conversion, Python 3.13 improvements, bug fixes
+
+---
+
 ## [0.10.0] - 2025-08-04
 
 # MFLUX v.0.10.0 Release Notes
diff --git a/Makefile b/Makefile
index 67e227e..be58b5c 100644
--- a/Makefile
+++ b/Makefile
@@ -93,7 +93,7 @@ check: ensure-ruff
 .PHONY: test
 test: ensure-pytest
 	# 🏗️ Running tests...
-	uv pip install mlx==0.27.1  # Install pinned MLX version specifically for testing
+	uv pip install mlx==0.29.2  # Install pinned MLX version specifically for testing
 	$(PYTHON) -m pytest
 	# ✅ Tests completed
 
diff --git a/README.md b/README.md
index ed790ee..cf0a0db 100644
--- a/README.md
+++ b/README.md
@@ -56,6 +56,7 @@ All models are implemented from scratch in MLX and only the tokenizers are used
 [Huggingface Transformers](https://github.com/huggingface/transformers) library. Other than that, there are only minimal dependencies
 like [Numpy](https://numpy.org) and [Pillow](https://pypi.org/project/pillow/) for simple image post-processing.
 
+As of v.0.11.0, MFLUX now supports the Qwen Image model.
 ---
 
 ### 💿 Installation
@@ -152,6 +153,12 @@ This example uses the more powerful `dev` model with 25 time steps:
 mflux-generate --model dev --prompt "Luxury food photograph" --steps 25 --seed 2 -q 8
 ```
 
+This example uses the `qwen` model with 20 time steps:
+
+```sh
+mflux-generate --model qwen --prompt "Luxury food photograph" --steps 20 --seed 2 -q 6
+```
+
 You can also pipe prompts from other commands using stdin:
 
 ```sh
@@ -168,8 +175,8 @@ from mflux.config.config import Config
 
 # Load the model
 flux = Flux1.from_name(
-   model_name="schnell",  # "schnell", "dev", or "krea-dev"
-       quantize=8,            # 3, 4, 5, 6, or 8
+   model_name="schnell",  # "schnell", "dev", "krea-dev", or "qwen"
+   quantize=8,            # 3, 4, 5, 6, or 8
 )
 
 # Generate an image
@@ -177,7 +184,7 @@ image = flux.generate_image(
    seed=2,
    prompt="Luxury food photograph",
    config=Config(
-      num_inference_steps=2,  # "schnell" works well with 2-4 steps, "dev" and "krea-dev" work well with 20-25 steps
+      num_inference_steps=2,  # "schnell" works well with 2-4 steps, "dev", "krea-dev", and "qwen" work well with 20-25 steps
       height=1024,
       width=1024,
    )
@@ -188,7 +195,7 @@ image.save(path="image.png")
 
 For more advanced Python usage and additional configuration options, you can explore the entry point files in the source code, such as [`generate.py`](src/mflux/generate.py), [`generate_controlnet.py`](src/mflux/generate_controlnet.py), [`generate_fill.py`](src/mflux/generate_fill.py), and others in the [`src/mflux/`](src/mflux/) directory. These files demonstrate how to use the Python API for various features and provide examples of advanced configurations.
 
-⚠️ *If the specific model is not already downloaded on your machine, it will start the download process and fetch the model weights (~34GB in size for the Schnell or Dev model respectively). See the [quantization](#%EF%B8%8F-quantization) section for running compressed versions of the model.* ⚠️
+⚠️ *If the specific model is not already downloaded on your machine, it will start the download process and fetch the model weights (~34GB for Schnell/Dev models, ~58GB for Qwen). See the [quantization](#%EF%B8%8F-quantization) section for running compressed versions of the model.* ⚠️
 
 *By default, mflux caches files in `~/Library/Caches/mflux/`. The Hugging Face model files themselves are cached separately in the Hugging Face cache directory (e.g., `~/.cache/huggingface/`).*
 
@@ -228,9 +235,9 @@ mflux-generate \
 
 - **`--prompt`** (required, `str`): Text description of the image to generate. Use `-` to read the prompt from stdin (e.g., `echo "A beautiful sunset" | mflux-generate --prompt -`).
 
-- **`--model`** or **`-m`** (required, `str`): Model to use for generation. Can be one of the official models (`"schnell"`, `"dev"`, or `"krea-dev"`) or a HuggingFace repository ID for a compatible third-party model (e.g., `"Freepik/flux.1-lite-8B-alpha"`).
+- **`--model`** or **`-m`** (required, `str`): Model to use for generation. Can be one of the official models (`"schnell"`, `"dev"`, `"krea-dev"`, or `"qwen"`) or a HuggingFace repository ID for a compatible third-party model (e.g., `"Freepik/flux.1-lite-8B-alpha"`).
 
-- **`--base-model`** (optional, `str`, default: `None`): Specifies which base architecture a third-party model is derived from (`"schnell"`, `"dev"`, or `"krea-dev"`). Required when using third-party models from HuggingFace.
+- **`--base-model`** (optional, `str`, default: `None`): Specifies which base architecture a third-party model is derived from (`"schnell"`, `"dev"`, `"krea-dev"`, or `"qwen"`). Required when using third-party models from HuggingFace.
 
 - **`--output`** (optional, `str`, default: `"image.png"`): Output image filename. If `--seed` or `--auto-seeds` establishes multiple seed values, the output filename will automatically be modified to include the seed value (e.g., `image_seed_42.png`).
 
@@ -244,7 +251,7 @@ mflux-generate \
 
 - **`--steps`** (optional, `int`, default: `4`): Number of inference steps.
 
-- **`--guidance`** (optional, `float`, default: `3.5`): Guidance scale (only used for `"dev"` model).
+- **`--guidance`** (optional, `float`, default: `3.5`): Guidance scale (only used for `"dev"`, `"krea-dev"` and `qwen` models).
 
 - **`--path`** (optional, `str`, default: `None`): Path to a local model on disk.
 
@@ -745,6 +752,8 @@ mflux-save \
     --quantize 8
 ```
 
+The `mflux-save` command works with both Flux and Qwen models.
+
 *Note that when saving a quantized version, you will need the original huggingface weights.*
 
 It is also possible to specify [LoRA](#-lora) adapters when saving the model, e.g
@@ -792,6 +801,7 @@ In other words, you can reclaim the 34GB diskspace (per model) by deleting the f
   - [dhairyashil/FLUX.1-dev-mflux-4bit](https://huggingface.co/dhairyashil/FLUX.1-dev-mflux-4bit)
   - [akx/FLUX.1-Kontext-dev-mflux-4bit](https://huggingface.co/akx/FLUX.1-Kontext-dev-mflux-4bit)
   - [filipstrand/FLUX.1-Krea-dev-mflux-4bit](https://huggingface.co/filipstrand/FLUX.1-Krea-dev-mflux-4bit)
+  - [filipstrand/Qwen-Image-mflux-6bit](https://huggingface.co/filipstrand/Qwen-Image-mflux-6bit)
 
 
 Using the [community model support](#-third-party-huggingface-model-support), the quantized weights can be also be automatically downloaded when running the generate command:
@@ -940,7 +950,7 @@ mflux-generate --prompt "pikachu, Paper Cutout Style" --model schnell --steps 4
 
 *Note that LoRA trained weights are typically trained with a **trigger word or phrase**. For example, in the latter case, the sentence should include the phrase **"Paper Cutout Style"**.*
 
-*Also note that the same LoRA weights can work well with both the `schnell` and `dev` models. Refer to the original LoRA repository to see what mode it was trained for.*
+*Also note that the same LoRA weights can work well with both the `schnell` and `dev` models. Qwen models support LoRA but may require Qwen-specific LoRA weights. Refer to the original LoRA repository to see what mode it was trained for.*
 
 #### Multi-LoRA
 
@@ -1863,6 +1873,7 @@ See `uv run tools/rename_images.py --help` for full CLI usage help.
 - Set up shell aliases for required args examples:
   - shortcut for dev model: `alias mflux-dev='mflux-generate --model dev'`
   - shortcut for schnell model *and* always save metadata: `alias mflux-schnell='mflux-generate --model schnell --metadata'`
+  - shortcut for qwen model: `alias mflux-qwen='mflux-generate --model qwen'`
 - For systems with limited memory, use the `--low-ram` flag to reduce memory usage by constraining the MLX cache size and releasing components after use
 - On battery-powered Macs, use `--battery-percentage-stop-limit` (or `-B`) to prevent your laptop from shutting down during long generation sessions
 - When generating multiple images with different seeds, use `--seed` with multiple values or `--auto-seeds` to automatically generate a series of random seeds
diff --git a/pyproject.toml b/pyproject.toml
index 987f132..3e34f9c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -17,7 +17,7 @@ source-exclude = [
 
 [project]
 name = "mflux"
-version = "0.10.0"
+version = "0.11.0"
 description = "A MLX port of FLUX based on the Huggingface Diffusers implementation."
 readme = "README.md"
 keywords = ["diffusers", "flux", "mlx"]
@@ -30,7 +30,7 @@ dependencies = [
     "filelock>=3.18.0",
     "huggingface-hub>=0.24.5,<1.0",
     "matplotlib>=3.9.2,<4.0",
-    "mlx>=0.27.0,<0.28.0",
+    "mlx>=0.27.0,<0.30.0",
     "numpy>=2.0.1,<3.0",
     "opencv-python>=4.10.0,<5.0",
     "piexif>=1.1.3,<2.0",
@@ -68,7 +68,7 @@ dev = [
     "matplotlib>3.10,<4.0",
     "pytest>=8.3.0,<9.0",
     "pytest-timer>=1.0,<2.0",
-    "mlx==0.27.1",  # Used ONLY during test runs to ensure deterministic test results
+    "mlx==0.29.2",  # Used ONLY during test runs to ensure deterministic test results
 ]
 
 [project.urls]
diff --git a/src/mflux/config/model_config.py b/src/mflux/config/model_config.py
index 092122e..87e7e90 100644
--- a/src/mflux/config/model_config.py
+++ b/src/mflux/config/model_config.py
@@ -128,7 +128,9 @@ class ModelConfig:
             default_base = next((b for b in base_models if base_model == b.model_name or base_model in b.aliases), None)
         else:
             # Infer from model_name substring - prefer longer matches (more specific)
-            matching_bases = [(b, alias) for b in base_models for alias in b.aliases if alias and alias in model_name]
+            # Use case-insensitive matching for better compatibility
+            model_name_lower = model_name.lower()
+            matching_bases = [(b, alias) for b in base_models for alias in b.aliases if alias and alias.lower() in model_name_lower]
 
             if matching_bases:
                 # Sort by alias length descending, then by priority ascending
diff --git a/src/mflux/info.py b/src/mflux/info.py
index d88b37b..591a291 100644
--- a/src/mflux/info.py
+++ b/src/mflux/info.py
@@ -56,7 +56,7 @@ def format_metadata(metadata: dict) -> str:
     if lora_paths := exif.get("lora_paths"):
         lines.append("")
         lines.append(f"LoRAs ({len(lora_paths)}):")
-        lora_scales = exif.get("lora_scales", [])
+        lora_scales = exif.get("lora_scales") or []
         for i, lora in enumerate(lora_paths):
             scale = lora_scales[i] if i < len(lora_scales) else 1.0
             lora_name = Path(lora).name
diff --git a/src/mflux/models/common/lora/mapping/lora_loader.py b/src/mflux/models/common/lora/mapping/lora_loader.py
index 9632447..4f15826 100644
--- a/src/mflux/models/common/lora/mapping/lora_loader.py
+++ b/src/mflux/models/common/lora/mapping/lora_loader.py
@@ -174,33 +174,49 @@ class LoRALoader:
         effective_scale = scale
 
         # Create new LoRA layer
-        if hasattr(current_module, 'weight'):
-            # Create LoRA layer
-            lora_layer = LoRALinear.from_linear(
-                current_module,
-                r=lora_A.shape[1],
-                scale=effective_scale
-            )
-
-            # Set the LoRA matrices - use the correct dimensions from the LoRA file
-            lora_layer.lora_A = lora_A
-            lora_layer.lora_B = lora_B
-
-            # Apply alpha scaling to the matrices if present
-            if "alpha" in lora_data:
-                lora_layer.lora_B = lora_layer.lora_B * alpha_scale
-
+        # Check if it's a linear layer (either nn.Linear, LoRALinear, or FusedLoRALinear)
+        is_linear = hasattr(current_module, 'weight')
+        is_lora_linear = isinstance(current_module, LoRALinear)
+        is_fused_linear = isinstance(current_module, FusedLoRALinear)
+        
+        if is_linear or is_lora_linear or is_fused_linear:
             # Handle fusion: if the current module is already a LoRA layer, fuse them
-            if isinstance(current_module, LoRALinear):
+            if is_lora_linear:
                 print(f"   🔀 Fusing with existing LoRA at {target_path}")
+                # Create a temporary LoRA layer from the base linear of the existing LoRA
+                lora_layer = LoRALinear.from_linear(
+                    current_module.linear,
+                    r=lora_A.shape[1],
+                    scale=effective_scale
+                )
+                # Set the LoRA matrices
+                lora_layer.lora_A = lora_A
+                lora_layer.lora_B = lora_B
+                # Apply alpha scaling to the matrices if present
+                if "alpha" in lora_data:
+                    lora_layer.lora_B = lora_layer.lora_B * alpha_scale
+                
                 # Create fused layer with the existing LoRA and the new one
                 fused_layer = FusedLoRALinear(
                     base_linear=current_module.linear,
                     loras=[current_module, lora_layer]
                 )
                 replacement_layer = fused_layer
-            elif isinstance(current_module, FusedLoRALinear):
+            elif is_fused_linear:
                 print(f"   🔀 Adding to existing fusion at {target_path}")
+                # Create a temporary LoRA layer from the base linear
+                lora_layer = LoRALinear.from_linear(
+                    current_module.base_linear,
+                    r=lora_A.shape[1],
+                    scale=effective_scale
+                )
+                # Set the LoRA matrices
+                lora_layer.lora_A = lora_A
+                lora_layer.lora_B = lora_B
+                # Apply alpha scaling to the matrices if present
+                if "alpha" in lora_data:
+                    lora_layer.lora_B = lora_layer.lora_B * alpha_scale
+                
                 # Add to existing fusion
                 fused_layer = FusedLoRALinear(
                     base_linear=current_module.base_linear,
@@ -209,6 +225,19 @@ class LoRALoader:
                 replacement_layer = fused_layer
             else:
                 # First LoRA on this layer
+                # Create LoRA layer
+                lora_layer = LoRALinear.from_linear(
+                    current_module,
+                    r=lora_A.shape[1],
+                    scale=effective_scale
+                )
+                # Set the LoRA matrices - use the correct dimensions from the LoRA file
+                lora_layer.lora_A = lora_A
+                lora_layer.lora_B = lora_B
+                # Apply alpha scaling to the matrices if present
+                if "alpha" in lora_data:
+                    lora_layer.lora_B = lora_layer.lora_B * alpha_scale
+                
                 replacement_layer = lora_layer
 
             # Replace the layer in the parent module
diff --git a/src/mflux/models/qwen/variants/txt2img/qwen_image.py b/src/mflux/models/qwen/variants/txt2img/qwen_image.py
index 0d681f1..546fde5 100644
--- a/src/mflux/models/qwen/variants/txt2img/qwen_image.py
+++ b/src/mflux/models/qwen/variants/txt2img/qwen_image.py
@@ -13,6 +13,7 @@ from mflux.models.qwen.model.qwen_text_encoder.qwen_text_encoder import QwenText
 from mflux.models.qwen.model.qwen_transformer.qwen_transformer import QwenTransformer
 from mflux.models.qwen.model.qwen_vae.qwen_vae import QwenVAE
 from mflux.models.qwen.qwen_initializer import QwenImageInitializer
+from mflux.models.qwen.weights.qwen_model_saver import QwenModelSaver
 from mflux.post_processing.array_util import ArrayUtil
 from mflux.post_processing.generated_image import GeneratedImage
 from mflux.post_processing.image_util import ImageUtil
@@ -168,6 +169,9 @@ class QwenImage(nn.Module):
             negative_prompt=negative_prompt,
         )
 
+    def save_model(self, base_path: str) -> None:
+        QwenModelSaver.save_model(self, self.bits, base_path)
+
     @staticmethod
     def _compute_guided_noise(
         noise: mx.array,
diff --git a/src/mflux/models/qwen/weights/qwen_model_saver.py b/src/mflux/models/qwen/weights/qwen_model_saver.py
new file mode 100644
index 0000000..be98a63
--- /dev/null
+++ b/src/mflux/models/qwen/weights/qwen_model_saver.py
@@ -0,0 +1,56 @@
+from pathlib import Path
+
+import mlx.core as mx
+from mlx import nn
+from mlx.utils import tree_flatten
+from transformers import Qwen2Tokenizer
+
+from mflux.utils.version_util import VersionUtil
+
+
+class QwenModelSaver:
+    @staticmethod
+    def save_model(model, bits: int, base_path: str):
+        # Save the tokenizer
+        QwenModelSaver._save_tokenizer(base_path, model.qwen_tokenizer.tokenizer, "tokenizer")
+
+        # Save the models
+        QwenModelSaver.save_weights(base_path, bits, model.vae, "vae")
+        QwenModelSaver.save_weights(base_path, bits, model.transformer, "transformer")
+        QwenModelSaver.save_weights(base_path, bits, model.text_encoder, "text_encoder")
+
+    @staticmethod
+    def _save_tokenizer(base_path: str, tokenizer: Qwen2Tokenizer, subdir: str):
+        path = Path(base_path) / subdir
+        tokenizer.save_pretrained(path)
+
+    @staticmethod
+    def save_weights(base_path: str, bits: int, model: nn.Module, subdir: str):
+        path = Path(base_path) / subdir
+        path.mkdir(parents=True, exist_ok=True)
+        weights = QwenModelSaver._split_weights(base_path, dict(tree_flatten(model.parameters())))
+        for i, weight in enumerate(weights):
+            mx.save_safetensors(
+                str(path / f"{i}.safetensors"),
+                weight,
+                {
+                    "quantization_level": str(bits),
+                    "mflux_version": VersionUtil.get_mflux_version(),
+                },
+            )
+
+    @staticmethod
+    def _split_weights(base_path: str, weights: dict, max_file_size_gb: int = 2) -> list:
+        # Copied from mlx-examples repo
+        max_file_size_bytes = max_file_size_gb << 30
+        shards = []
+        shard, shard_size = {}, 0
+        for k, v in weights.items():
+            if shard_size + v.nbytes > max_file_size_bytes:
+                shards.append(shard)
+                shard, shard_size = {}, 0
+            shard[k] = v
+            shard_size += v.nbytes
+        shards.append(shard)
+        return shards
+
diff --git a/src/mflux/models/qwen/weights/qwen_weight_handler.py b/src/mflux/models/qwen/weights/qwen_weight_handler.py
index b70e04d..39b3235 100644
--- a/src/mflux/models/qwen/weights/qwen_weight_handler.py
+++ b/src/mflux/models/qwen/weights/qwen_weight_handler.py
@@ -56,18 +56,80 @@ class QwenWeightHandler:
     @staticmethod
     def load_transformer(root_path: Path) -> tuple[dict, int | None, str | None]:
         flat = QwenWeightHandler._load_safetensors_shards(root_path / "transformer", loading_mode="multi_glob")
+
+        # Check if this is a saved quantized model (with metadata) or HuggingFace weights
+        # Saved models from mflux-save are already in MLX structure and don't need manual mapping
+        quantization_level = None
+        mflux_version = None
+
+        # Try to get metadata from the first weight shard
+        import mlx.core as mx
+        from mlx.utils import tree_unflatten
+
+        file_glob = sorted((root_path / "transformer").glob("*.safetensors"))
+        if file_glob:
+            data = mx.load(str(file_glob[0]), return_metadata=True)
+            if len(data) > 1:
+                quantization_level = data[1].get("quantization_level")
+                mflux_version = data[1].get("mflux_version")
+
+        # If this is a saved model (has metadata), use tree_unflatten directly
+        if quantization_level is not None or mflux_version is not None:
+            return tree_unflatten(list(flat.items())), quantization_level, mflux_version
+
+        # Otherwise, it's HuggingFace weights that need manual mapping
         mapped_weights = QwenWeightHandler._manual_transformer_mapping(flat)
         return mapped_weights, None, None
 
     @staticmethod
     def _load_qwen_text_encoder(root_path: Path) -> tuple[dict, int | None, str | None]:
+        import mlx.core as mx
+        from mlx.utils import tree_unflatten
+
+        # Check for saved model metadata FIRST to determine loading mode
+        quantization_level = None
+        mflux_version = None
+        file_glob = sorted((root_path / "text_encoder").glob("*.safetensors"))
+        if file_glob:
+            data = mx.load(str(file_glob[0]), return_metadata=True)
+            if len(data) > 1:
+                quantization_level = data[1].get("quantization_level")
+                mflux_version = data[1].get("mflux_version")
+
+        # If this is a saved model, load without expecting index.json
+        if quantization_level is not None or mflux_version is not None:
+            all_weights = QwenWeightHandler._load_safetensors_shards(
+                root_path / "text_encoder", loading_mode="multi_glob"
+            )
+            return tree_unflatten(list(all_weights.items())), quantization_level, mflux_version
+
+        # Otherwise, it's HuggingFace weights that need manual mapping
         all_weights = QwenWeightHandler._load_safetensors_shards(root_path / "text_encoder", loading_mode="multi_json")
         mapped_weights = QwenWeightHandler._manual_text_encoder_mapping(all_weights)
         return mapped_weights, None, None
 
     @staticmethod
     def _load_vae(root_path: Path) -> tuple[dict, int | None, str | None]:
+        import mlx.core as mx
+        from mlx.utils import tree_unflatten
+
         weights = QwenWeightHandler._load_safetensors_shards(root_path / "vae", loading_mode="single")
+
+        # Check for saved model metadata
+        quantization_level = None
+        mflux_version = None
+        file_glob = sorted((root_path / "vae").glob("*.safetensors"))
+        if file_glob:
+            data = mx.load(str(file_glob[0]), return_metadata=True)
+            if len(data) > 1:
+                quantization_level = data[1].get("quantization_level")
+                mflux_version = data[1].get("mflux_version")
+
+        # If this is a saved model, use tree_unflatten directly
+        if quantization_level is not None or mflux_version is not None:
+            return tree_unflatten(list(weights.items())), quantization_level, mflux_version
+
+        # Otherwise, it's HuggingFace weights that need manual mapping and reshaping
         reshaped_weights = [QwenWeightUtil.reshape_weights(k, v) for k, v in weights.items()]
         reshaped_weights = QwenWeightUtil.flatten(reshaped_weights)
         weights = dict(reshaped_weights)
diff --git a/src/mflux/post_processing/metadata_builder.py b/src/mflux/post_processing/metadata_builder.py
index a0dffbd..9818f86 100644
--- a/src/mflux/post_processing/metadata_builder.py
+++ b/src/mflux/post_processing/metadata_builder.py
@@ -214,7 +214,8 @@ class MetadataBuilder:
         Returns:
             Comma-separated string of LoRA names and scales, or empty string
         """
-        if "lora_paths" not in metadata or not metadata["lora_paths"]:
+        # Check if lora_paths exists and is not None/empty
+        if "lora_paths" not in metadata or metadata["lora_paths"] is None or not metadata["lora_paths"]:
             return ""
 
         lora_list = []
diff --git a/src/mflux/save.py b/src/mflux/save.py
index 13790b4..2c6ab7d 100644
--- a/src/mflux/save.py
+++ b/src/mflux/save.py
@@ -1,23 +1,27 @@
 from mflux.config.model_config import ModelConfig
 from mflux.models.flux.variants.txt2img.flux import Flux1
+from mflux.models.qwen.variants.txt2img.qwen_image import QwenImage
 from mflux.ui.cli.parsers import CommandLineParser
 
 
 def main():
     # 0. Parse command line arguments
-    parser = CommandLineParser(description="Save a quantized version of Flux.1 to disk.")  # fmt: off
+    parser = CommandLineParser(description="Save a quantized version of a model to disk.")  # fmt: off
     parser.add_model_arguments(path_type="save", require_model_arg=True)
     parser.add_lora_arguments()
     args = parser.parse_args()
 
-    # 1. Load, quantize and save the model
-    flux = Flux1(
+    # 1. Determine model class based on model name
+    model_class = QwenImage if "qwen" in args.model.lower() else Flux1
+
+    # 2. Load, quantize and save the model
+    model = model_class(
         model_config=ModelConfig.from_name(args.model, base_model=args.base_model),
         quantize=args.quantize,
         lora_paths=args.lora_paths,
         lora_scales=args.lora_scales,
     )
-    flux.save_model(args.path)
+    model.save_model(args.path)
 
 
 if __name__ == "__main__":
diff --git a/tests/resources/reference_controlnet_dev_lora_depth.png b/tests/resources/reference_controlnet_dev_lora_depth.png
index 8ea75a6..0cd2da4 100644
Binary files a/tests/resources/reference_controlnet_dev_lora_depth.png and b/tests/resources/reference_controlnet_dev_lora_depth.png differ
diff --git a/uv.lock b/uv.lock
index 76bb96b..2b00033 100644
--- a/uv.lock
+++ b/uv.lock
@@ -755,7 +755,7 @@ wheels = [
 
 [[package]]
 name = "mflux"
-version = "0.10.0"
+version = "0.11.0"
 source = { editable = "." }
 dependencies = [
     { name = "accelerate" },
@@ -799,8 +799,8 @@ requires-dist = [
     { name = "huggingface-hub", specifier = ">=0.24.5,<1.0" },
     { name = "matplotlib", specifier = ">=3.9.2,<4.0" },
     { name = "matplotlib", marker = "extra == 'dev'", specifier = ">3.10,<4.0" },
-    { name = "mlx", specifier = ">=0.27.0,<0.28.0" },
-    { name = "mlx", marker = "extra == 'dev'", specifier = "==0.27.1" },
+    { name = "mlx", specifier = ">=0.27.0,<0.30.0" },
+    { name = "mlx", marker = "extra == 'dev'", specifier = "==0.29.2" },
     { name = "numpy", specifier = ">=2.0.1,<3.0" },
     { name = "opencv-python", specifier = ">=4.10.0,<5.0" },
     { name = "piexif", specifier = ">=1.1.3,<2.0" },
@@ -827,38 +827,38 @@ requires-dist = [
 
 [[package]]
 name = "mlx"
-version = "0.27.1"
+version = "0.29.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "mlx-metal", marker = "platform_system == 'Darwin'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/c9/ed/07a7145c500914ed8ab704a7a3828a5134e85f8b4458e1fee887306867d6/mlx-0.27.1-cp310-cp310-macosx_13_0_arm64.whl", hash = "sha256:a033b65fe46425ad5032867d5a71a556a5168108d89aa7092b457556c70d84fc", size = 557679 },
-    { url = "https://files.pythonhosted.org/packages/aa/ae/b3e46904ea04b5badb77195169880733f09976a3dfc81c954b99f9adbbc4/mlx-0.27.1-cp310-cp310-macosx_14_0_arm64.whl", hash = "sha256:2836c6fd9803dc0c6cd06f204e31b3e0191e6c5b6bc8570b28661d926908cba3", size = 530556 },
-    { url = "https://files.pythonhosted.org/packages/6e/24/85f5a7dfcf0b4974aa43a2f6dc0d59dfc04e4a1051aacebd7e7810b823ff/mlx-0.27.1-cp310-cp310-macosx_15_0_arm64.whl", hash = "sha256:8b0566054c46d84c470cc99cda2afc3914ad6c7808fbb724dc1ec235e5b2a98c", size = 530559 },
-    { url = "https://files.pythonhosted.org/packages/14/bd/91ad900837a3760ba056863c1350f0deabc3e462c3999b4d5ea875368177/mlx-0.27.1-cp310-cp310-manylinux_2_35_x86_64.whl", hash = "sha256:91ef93ce09900c9a8ca662cf34e3c39ab5af2762822ecd6b12fecae518be167f", size = 628183 },
-    { url = "https://files.pythonhosted.org/packages/ac/63/88cb9a5019bea21244385eacc4c64deb4d074a8bc3b4b6d2d97cdacf97a2/mlx-0.27.1-cp311-cp311-macosx_13_0_arm64.whl", hash = "sha256:d2e5dedbfbcbe558e51a5c476ca6a18e307676f9e49854eb27e53778bc474699", size = 557582 },
-    { url = "https://files.pythonhosted.org/packages/fa/87/c2d7343e7b054481c52e23a190911c40aed4f8c77a8dfeda1f48d5cb520a/mlx-0.27.1-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:9f04b9778897a879c9ca22e5413dfa1efc192d86d7211b184e079efec49dfb8b", size = 530820 },
-    { url = "https://files.pythonhosted.org/packages/06/bf/20497ca7411028baa56372c20e94a3eaddac973155b415f08fc12592c2cf/mlx-0.27.1-cp311-cp311-macosx_15_0_arm64.whl", hash = "sha256:01d794f9e390438ab4f942a18d9a8ca65bef10c2c2007ef38ca988d039d6d9d3", size = 530821 },
-    { url = "https://files.pythonhosted.org/packages/48/68/234a0d846576502ea1bf8ea478c764d87eec9042349e298d2aea58c4e8e9/mlx-0.27.1-cp311-cp311-manylinux_2_35_x86_64.whl", hash = "sha256:fae11432d0639789f1e172b19b35ac8987c8ab9716e55a23fc7a170d6545fc33", size = 628500 },
-    { url = "https://files.pythonhosted.org/packages/65/43/125102bbb2be6825880ae2dc8d8702f99cfa7753407f574457b36e422218/mlx-0.27.1-cp312-cp312-macosx_13_0_arm64.whl", hash = "sha256:0c570c9afb57c697bd864504115be8a7c4de97f0b80557a597d496ee426a6812", size = 549869 },
-    { url = "https://files.pythonhosted.org/packages/f3/79/0bf681700fc8b165517e907f9ec777b5a5d628004a65a777148f68c6baa0/mlx-0.27.1-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:ccff7bbbd9df302b26e79013ef6d0c3531c9ba5963ead521e2d85856811b86a0", size = 531671 },
-    { url = "https://files.pythonhosted.org/packages/ec/97/f1367b4892bef7f78e38737d3a28094e93124f11684a28a9e92ed5a13b2b/mlx-0.27.1-cp312-cp312-macosx_15_0_arm64.whl", hash = "sha256:9ccadaed449c07dfeae484620992b904c17dfea7564f8df63095c60eed3af02b", size = 531672 },
-    { url = "https://files.pythonhosted.org/packages/9c/0f/3ebec0806ed2c5ba8ba151394cd62a46b4d83ea347da33af91b86d8969d4/mlx-0.27.1-cp312-cp312-manylinux_2_35_x86_64.whl", hash = "sha256:742c413e75605b71db69379a176da63e32ba19b9e9ad03b8763adbd1fcfcd394", size = 621661 },
-    { url = "https://files.pythonhosted.org/packages/86/f6/4324386b0764deb692e14a97282a348a9a938aa8b441bf8b6c7599f418d4/mlx-0.27.1-cp313-cp313-macosx_13_0_arm64.whl", hash = "sha256:803669a28031766c2b0fe94c0a3bfd030184e706092f0a831b33620c1e2ef865", size = 549847 },
-    { url = "https://files.pythonhosted.org/packages/cf/4b/3194ccb03527a050c04d837d731a11599f8620e6ce16d3971798caae1d44/mlx-0.27.1-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:e9649b3d86ce564000797384510c9d07af38a9ce2a07df8e2f7c6a3e0f0f059e", size = 531664 },
-    { url = "https://files.pythonhosted.org/packages/cc/57/a6e0d8dc6e7ba08a64d71fb89d743e77446040113ea1dbb7950be8f60f39/mlx-0.27.1-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:5c501ceec7c6aa2ea1c850dd4e14e679f5416a142264b9d5d405a4e0aeb991b2", size = 531663 },
-    { url = "https://files.pythonhosted.org/packages/48/96/6932e0fa866d08cdf77a3fc37cf624e4cac8960d2e297c3fd80afb49417d/mlx-0.27.1-cp313-cp313-manylinux_2_35_x86_64.whl", hash = "sha256:42bfbabdcd0b7aa1c94748c8963e66680a3cf3f599b01036db14e496bf276eb8", size = 621609 },
+    { url = "https://files.pythonhosted.org/packages/3f/f0/2c2f99a91ed9dfcc78d31d9e5d3bb2f5305a8d65953cbc41f34f8056c49a/mlx-0.29.2-cp310-cp310-macosx_13_0_arm64.whl", hash = "sha256:b46c1a24b9b8f7145e4d84410552ddfa03f40f9afdbe8f819f6b4b52b4db5d30", size = 547369 },
+    { url = "https://files.pythonhosted.org/packages/3b/06/0edddf0a5facb58c17616cd33da6de2722636da8d8d3927272a5a88658e4/mlx-0.29.2-cp310-cp310-macosx_14_0_arm64.whl", hash = "sha256:18c5b63c6e4b35f75f477f74d3942870e0bc9c9f1c6a071d8a058bbd46681bf4", size = 547367 },
+    { url = "https://files.pythonhosted.org/packages/e1/cd/8c089cf1678a752ed4175a7ad5f08823ceef75d797217e12a44a71d6c062/mlx-0.29.2-cp310-cp310-macosx_15_0_arm64.whl", hash = "sha256:ed53b8383ad4ee4311400558e0a5ec61105fc4553950b5732c66e7081cd1a9e8", size = 547365 },
+    { url = "https://files.pythonhosted.org/packages/8b/cc/9a14ac57a251f0ee5047f42214fa50d1d5223d805f0b8b6627639c3692eb/mlx-0.29.2-cp310-cp310-manylinux_2_35_x86_64.whl", hash = "sha256:fff27af8dca74546422e8400d32ab848baf42405123cbcfdf62674a9c0560ebe", size = 652302 },
+    { url = "https://files.pythonhosted.org/packages/cb/f0/f57349f37cf5dd53f95127e141fc59fc435e4b6bfabba5a84c65de4d3597/mlx-0.29.2-cp311-cp311-macosx_13_0_arm64.whl", hash = "sha256:e74965369227230374b3e8e8c8d46e209e5221a9b76bbb0fa788617e2c68f73c", size = 547581 },
+    { url = "https://files.pythonhosted.org/packages/66/04/e016ca28dc9e0738a2541581420125cfe6bba24466a64420600bdd6fd52c/mlx-0.29.2-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:0f79194eeac78e85b96439d3bbc17aae5aba045a2af083c000b4fbbc501f253e", size = 547581 },
+    { url = "https://files.pythonhosted.org/packages/1f/b3/e2595e70ef8d4438dff694857745b0e108911e5b5fb83259dde6e5dc5bd1/mlx-0.29.2-cp311-cp311-macosx_15_0_arm64.whl", hash = "sha256:33bbbb0fd24895d5ff080bb4d10e3e77017bba675d9a12466c8866eaf9b47854", size = 547578 },
+    { url = "https://files.pythonhosted.org/packages/11/8c/5d51543ab128c2dff5e4b44ca799db8db5aa4f4ffc34af6531fd73627b54/mlx-0.29.2-cp311-cp311-manylinux_2_35_x86_64.whl", hash = "sha256:32e159f2772be893bec580d2d50c0e6b32ad71a19ded7307bf6c871c8aaa9cf2", size = 651900 },
+    { url = "https://files.pythonhosted.org/packages/f3/84/7250237039e91d8e44ca0cf3522f189164844c196f262509afd29ef54710/mlx-0.29.2-cp312-cp312-macosx_13_0_arm64.whl", hash = "sha256:eec950bf7118ad0865d0fc4686bd85d99bf8463fc717d836a5132e1a08b4f129", size = 548336 },
+    { url = "https://files.pythonhosted.org/packages/13/47/428ac8d9b0cb5c136e5ce6c726cfdd55caa5b9497dafb6221acfee18f145/mlx-0.29.2-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:bef7333268d6d02e50a9ac6b10f661b711cd02da4a5e2d7619cf198a7e530308", size = 548334 },
+    { url = "https://files.pythonhosted.org/packages/14/f0/7d5d3527ca3fdc664c900b4b822028691739e58c8e8f7975b33df4d3536e/mlx-0.29.2-cp312-cp312-macosx_15_0_arm64.whl", hash = "sha256:f622fc6a84542a08ad2136e9251822d2c08106e5a1a0bd5d249a2d72bccd6577", size = 548330 },
+    { url = "https://files.pythonhosted.org/packages/09/18/e202e0f6232822f6768995cdbf50eda202137bb6547368f6e3993dbee00b/mlx-0.29.2-cp312-cp312-manylinux_2_35_x86_64.whl", hash = "sha256:a1aa1aee8e1b6bd1e51361e6b692c70d281b8187b2e859e70ecc11daab306dac", size = 648728 },
+    { url = "https://files.pythonhosted.org/packages/a0/9a/91f6f5d031f109fa8c00ba9dd4f7a3fc42e1097a57c26783ce000069c264/mlx-0.29.2-cp313-cp313-macosx_13_0_arm64.whl", hash = "sha256:05ea54173f4bde11b2c93e673d65d72523f5d850f5112d3874156a6fc74ca591", size = 548297 },
+    { url = "https://files.pythonhosted.org/packages/2b/2d/dae7ca0b7fa68c6c1f2b896dfe1b8060647f144d5c5da2d53388e38809b1/mlx-0.29.2-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:199dd029b5e55b6d94f1ce366d0137824e46e4333891424dd00413c739f50ae9", size = 548305 },
+    { url = "https://files.pythonhosted.org/packages/b1/56/f02f5c9e1fc11c020982501a763fa92b497ea50671a587760543987ba8c8/mlx-0.29.2-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:b6dd4e5f227414882b1676d99250d99389228d1bdc14e4e4e88c95d4903810b7", size = 548302 },
+    { url = "https://files.pythonhosted.org/packages/4b/b4/b61eeb92c424947675492dec3a411bdbeae307dfd78162d65ab47e8c3b4f/mlx-0.29.2-cp313-cp313-manylinux_2_35_x86_64.whl", hash = "sha256:c3b9a9aee13f346d060966472954eebe99d9f1b295c9a237c9a000f1ef9adf2c", size = 648709 },
 ]
 
 [[package]]
 name = "mlx-metal"
-version = "0.27.1"
+version = "0.29.2"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/8b/77/89fa3327011f018638c9e943e1edc081ce9ca0ed296fe64d6daf93c6ff51/mlx_metal-0.27.1-py3-none-macosx_13_0_arm64.whl", hash = "sha256:c66d9b1adb3c0ea19492fba6493f672bc7542e65dd65f7e2995918815fbeb907", size = 33523035 },
-    { url = "https://files.pythonhosted.org/packages/d7/a8/ac706ad6ce834834762d5146d791f77710efc896c13ef47fd7d672099056/mlx_metal-0.27.1-py3-none-macosx_14_0_arm64.whl", hash = "sha256:fe4415ddd242974d91c7ca0699cd01507d17da8a5ba304122ef137cdb5e7fff4", size = 32926383 },
-    { url = "https://files.pythonhosted.org/packages/78/77/6963681fb54ecaa0ae5de4209c15504a803a0edd1a33fd074e6c558fd5e0/mlx_metal-0.27.1-py3-none-macosx_15_0_arm64.whl", hash = "sha256:d025dea30bda8baa32c928cfa333eac64a5adc8d07656f8fc55072d99403ebc9", size = 32897065 },
+    { url = "https://files.pythonhosted.org/packages/31/a5/a045006546fed791f6e9a74ed4451dac871d3c35f9e54a3a25d820668a85/mlx_metal-0.29.2-py3-none-macosx_13_0_arm64.whl", hash = "sha256:cf8f83a521e620357185c57945142718d526b9312ee112e5a89eb5600480f4d6", size = 35056194 },
+    { url = "https://files.pythonhosted.org/packages/4c/8c/4bdd3a7d04ed477b32aec30d30236dfca9f9ac27706cb309511278ddd281/mlx_metal-0.29.2-py3-none-macosx_14_0_arm64.whl", hash = "sha256:fa944001970813b296e8aff5616f2fa9daeda6bc1d190c17fbe8a7ca838ecef0", size = 34791708 },
+    { url = "https://files.pythonhosted.org/packages/b1/11/12e158848fe4d3316c999ffb6c2d88f554bde98d69022b3385e25ece997e/mlx_metal-0.29.2-py3-none-macosx_15_0_arm64.whl", hash = "sha256:08d8b7fe305425a14b74ebf36cee176575bfd4cd8d34a2aaae8f05b9983d2d71", size = 34784506 },
 ]
 
 [[package]]