Hugging Face's logo

ussoewwin
/
ControlNet-models-ConvRot-INT8

Image-to-Image
controlnet
text-to-image
inpainting
sdxl
stable-diffusion-xl
illustrious-xl
z-image
z-image-turbo
qwen-image
qwen-image-2512
flux
flux.1-dev
diffsynth
diffsynth-studio
quantized
int8
convrot
comfyui
Model card Files
xet
 Community
ussoewwin commited on
Commit
0acd63f
·
verified ·
1 Parent(s): 92a150d

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +69 -30
README.md CHANGED
@@ -25,70 +25,109 @@ pipeline_tag: image-to-image
25
 
26
  # ControlNet Models (ConvRot INT8)
27
 
 
 
 
 
28
  High-fidelity **ConvRot INT8** quantized weights for multi-condition and dedicated ControlNet models across diverse generative architectures (**Illustrious-XL / SDXL**, **SDXL 1.0**, **Z-Image-Turbo**, **Qwen-Image / Qwen-Image-2512**, and **FLUX.1-dev**).
29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  ---
31
 
32
  ## 🌟 Model Overview
33
 
34
- This repository hosts high-quality **ConvRot INT8** quantized weights for all-in-one ControlNet Union and dedicated condition-specific ControlNet models. By applying orthogonal Hadamard rotation prior to per-channel INT8 quantization, these models effectively eliminate activation outlier distortion and drastically reduce VRAM and disk footprint while preserving precise structural control fidelity:
35
 
36
  - **CN-anytest4_illustrious2 (Variants A & B)**: Multi-purpose all-in-one Anytest v4 ControlNet models fine-tuned for **Illustrious-XL / SDXL** by **2vXpSwA7**, offering high-precision anime and illustration structure guiding.
37
  - **controlnet-union-pro-max-sdxl-1.0**: The comprehensive all-in-one ControlNet Union model for **SDXL 1.0** by **xinsir**, supporting 10+ control conditions in a single compact file.
38
- - **Z-Image-Turbo-Fun-Controlnet-Union-2.1-lite-2601-8steps**: Multi-condition ControlNet Union model (lite 6-block architecture, 8-step distilled) for the **Z-Image-Turbo** architecture by **alibaba-pai**, providing lightweight and fast structure control (Canny, Depth, Pose, Color, etc.).
39
- - **Z-Image-Turbo-Fun-Controlnet-Tile-2.1-lite-2601-8steps**: Dedicated high-resolution Tile and upscaling ControlNet model (lite architecture, 8-step distilled) for the **Z-Image-Turbo** architecture by **alibaba-pai**, optimized for super-resolution detail enhancement.
40
  - **Qwen-Image-2512-Fun-Controlnet-Union-2602**: Multi-condition ControlNet Union model (5 layer blocks) for the **Qwen-Image-2512** architecture.
41
  - **Qwen-Image-ControlNet-Inpainting**: Dedicated inpainting and editing ControlNet model for the **Qwen-Image** architecture.
42
- - **Qwen-Image DiffSynth ControlNet Suite**: Dedicated single-condition ControlNet models for the **Qwen-Image** architecture developed by **DiffSynth-Studio**:
43
- - **`qwen_image_canny_diffsynth_controlnet_convrot_int8`**: High-precision Canny edge structural control.
44
- - **`qwen_image_depth_diffsynth_controlnet_convrot_int8`**: Geometric spatial depth guidance.
45
- - **`qwen_image_inpaint_diffsynth_controlnet_convrot_int8`**: Masked inpainting and localized semantic replacement.
46
  - **FLUX.1-dev-ControlNet-Union-Pro-2.0**: Next-generation unified 7-in-1 ControlNet for the **FLUX.1-dev** architecture by **Shakker Labs**.
47
 
48
  ---
49
 
50
  ## 📦 Available Models
51
 
52
- | Filename | Base Architecture | Base Model / Author | Supported Conditions | Quantization | File Size | License |
53
- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
54
- | `CN-anytest4_illustrious2_A_hswq_convrot_int8.safetensors` | Illustrious-XL / SDXL | [2vXpSwA7/iroiro-lora (Anytest v4 Variant A)](https://huggingface.co/2vXpSwA7/iroiro-lora) | Multi-condition (Canny, Lineart, Depth, Pose, Structure) | ConvRot INT8 | ~1.43 GB | Fair AI / OpenRAIL++-M |
55
- | `CN-anytest4_illustrious2_B_hswq_convrot_int8.safetensors` | Illustrious-XL / SDXL | [2vXpSwA7/iroiro-lora (Anytest v4 Variant B)](https://huggingface.co/2vXpSwA7/iroiro-lora) | Multi-condition (Canny, Lineart, Depth, Pose, Structure) | ConvRot INT8 | ~1.42 GB | Fair AI / OpenRAIL++-M |
56
- | `controlnet-union-pro-max-sdxl-1.0_hswq_convrot_int8.safetensors` | SDXL 1.0 | [xinsir/controlnet-union-sdxl-1.0](https://huggingface.co/xinsir/controlnet-union-sdxl-1.0) | OpenPose, Depth, Canny, Lineart, Anime Lineart, Scribble, Soft Edge, Normal, Segment, Tile, Inpaint | ConvRot INT8 | ~1.44 GB | OpenRAIL++-M / Apache-2.0 |
57
- | `Z-Image-Turbo-Fun-Controlnet-Union-2.1-lite-2601-8steps_convrot_int8.safetensors` | Z-Image-Turbo | [alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1](https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1) | Multi-condition (Canny, Depth, Pose, Color, Scribble, Inpaint) | ConvRot INT8 | ~1.01 GB | Apache-2.0 |
58
- | `Z-Image-Turbo-Fun-Controlnet-Tile-2.1-lite-2601-8steps_convrot_int8.safetensors` | Z-Image-Turbo | [alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1](https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1) | Tile, Super-Resolution, Detail Enhancing | ConvRot INT8 | ~1.01 GB | Apache-2.0 |
59
- | `Qwen-Image-2512-Fun-Controlnet-Union-2602_convrot_int8.safetensors` | Qwen-Image-2512 | [alibaba-pai/Qwen-Image-2512-Fun-Controlnet-Union](https://huggingface.co/alibaba-pai/Qwen-Image-2512-Fun-Controlnet-Union) | Canny, HED, Depth, Pose, MLSD, Scribble, Gray, Inpaint | ConvRot INT8 | ~1.76 GB | Apache-2.0 |
60
- | `Qwen-Image-ControlNet-Inpainting_convrot_int8.safetensors` | Qwen-Image | [alibaba-pai/Qwen-Image-ControlNet-Inpainting](https://huggingface.co/alibaba-pai/Qwen-Image-ControlNet-Inpainting) | Inpainting, Image Editing | ConvRot INT8 | ~2.12 GB | Apache-2.0 |
61
- | `qwen_image_canny_diffsynth_controlnet_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Canny Edge Detection & Structural Guidance | ConvRot INT8 | ~1.14 GB | Apache-2.0 |
62
- | `qwen_image_depth_diffsynth_controlnet_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Geometric Depth Map Guidance | ConvRot INT8 | ~1.14 GB | Apache-2.0 |
63
- | `qwen_image_inpaint_diffsynth_controlnet_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Masked Inpainting & Localized Editing | ConvRot INT8 | ~1.14 GB | Apache-2.0 |
64
- | `FLUX.1-dev-ControlNet-Union-Pro-2.0_hswq_convrot_int8.safetensors` | FLUX.1-dev | [Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro-2.0](https://huggingface.co/Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro-2.0) | Canny, Depth, Pose, Blur, Gray, Soft Edge, Low Quality | ConvRot INT8 | ~2.71 GB | Other / Non-Commercial |
 
 
65
 
66
  ---
67
 
68
- ## 🛠️ Key Features
69
 
70
- - **All-in-One & Dedicated Condition Control**: Unified multi-condition architectures alongside specialized single-condition models (Canny, Depth, Tile, Inpainting) for fine-grained workflow flexibility.
71
- - **ConvRot INT8 Precision**: Leverages orthogonal Hadamard rotations to redistribute channel-wise outlier spikes uniformly across dimensions, preventing quantization error buildup in deep control layers.
72
- - **VRAM & Storage Optimization**: Slashes VRAM consumption and disk footprint by ~50% compared to unquantized FP16 checkpoints, allowing seamless multi-ControlNet workflows on consumer GPUs.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73
 
74
  ---
75
 
76
- ## 🚀 Usage in ComfyUI
77
 
78
- To load and execute these ConvRot INT8 ControlNet models in ComfyUI, please use the dedicated loader node from the **ComfyUI-HSWQ-Loader-and-Tools** extension:
 
 
 
79
 
80
- - **Extension Repository:** [ComfyUI-HSWQ-Loader-and-Tools](https://github.com/ussoewwin/ComfyUI-HSWQ-Loader-and-Tools)
81
 
82
- ### Installation
83
 
84
- Clone the repository into your ComfyUI `custom_nodes` directory:
85
 
86
  ```bash
87
  cd ComfyUI/custom_nodes
88
  git clone https://github.com/ussoewwin/ComfyUI-HSWQ-Loader-and-Tools.git
89
  ```
90
 
91
- Place the downloaded `.safetensors` files into your ComfyUI `models/controlnet/` directory and load them using the dedicated ControlNet loader node.
 
 
 
92
 
93
  ---
94
 
 
25
 
26
  # ControlNet Models (ConvRot INT8)
27
 
28
+ <p align="center">
29
+ <img src="https://raw.githubusercontent.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/main/icon.png" width="128">
30
+ </p>
31
+
32
  High-fidelity **ConvRot INT8** quantized weights for multi-condition and dedicated ControlNet models across diverse generative architectures (**Illustrious-XL / SDXL**, **SDXL 1.0**, **Z-Image-Turbo**, **Qwen-Image / Qwen-Image-2512**, and **FLUX.1-dev**).
33
 
34
+ - **Technical Details & Quantization Framework:** [Hybrid-Sensitivity-Weighted-Quantization (HSWQ)](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization)
35
+ - **ComfyUI Loader for ConvRot INT8:** [ComfyUI-HSWQ-Loader-and-Tools](https://github.com/ussoewwin/ComfyUI-HSWQ-Loader-and-Tools)
36
+
37
+ ---
38
+
39
+ ## ⚠️ Where each pack goes, and how it loads
40
+
41
+ **This repository mixes two runtime object types.** They use different loaders, and getting this wrong means the file silently fails to apply.
42
+
43
+ | group | the runtime object | the loader | the directory |
44
+ | :--- | :--- | :--- | :--- |
45
+ | **ControlNet** (SDXL anytest A/B, SDXL union-pro-max, Qwen-Image-2512 Union 2602, Qwen-Image Inpainting, FLUX.1-dev Union Pro 2.0) | a ControlNet | `ControlNetLoader` (stock) or the HSWQ ControlNet loader | `ComfyUI/models/controlnet/` |
46
+ | **Model patch** (Z-Image-Turbo Fun Controlnet Union/Tile, Qwen-Image DiffSynth canny/depth/inpaint) | a **model patch**, despite the "ControlNet" name | **`HSWQ Model Patch Loader`** (`HSWQModelPatchLoaderCustom`). The **stock `ModelPatchLoader` cannot read the INT8 layout** | `ComfyUI/models/model_patches/` |
47
+
48
+ The model patches are applied by the dedicated apply nodes (`ZImageFunControlnet`, `QwenImageDiffsynthControlnet`), not by `Apply ControlNet`.
49
+
50
  ---
51
 
52
  ## 🌟 Model Overview
53
 
54
+ This repository hosts high-quality **ConvRot INT8** quantized weights for all-in-one ControlNet Union and dedicated condition-specific ControlNet models. Orthogonal Hadamard rotation is applied to each 2D weight before per-channel INT8 quantization, which suppresses activation outlier distortion and cuts the disk footprint roughly in half while preserving structural control fidelity:
55
 
56
  - **CN-anytest4_illustrious2 (Variants A & B)**: Multi-purpose all-in-one Anytest v4 ControlNet models fine-tuned for **Illustrious-XL / SDXL** by **2vXpSwA7**, offering high-precision anime and illustration structure guiding.
57
  - **controlnet-union-pro-max-sdxl-1.0**: The comprehensive all-in-one ControlNet Union model for **SDXL 1.0** by **xinsir**, supporting 10+ control conditions in a single compact file.
58
+ - **Z-Image-Turbo-Fun-Controlnet-Union-2.1-lite-2602-8steps**: Multi-condition ControlNet Union model (lite architecture, 8-step distilled) for the **Z-Image-Turbo** architecture by **alibaba-pai**, providing lightweight and fast structure control (Canny, Depth, Pose, Color, etc.). **Model patch.**
59
+ - **Z-Image-Turbo-Fun-Controlnet-Tile-2.1-lite-2601-8steps**: Dedicated high-resolution Tile and upscaling ControlNet model (lite architecture, 8-step distilled) for the **Z-Image-Turbo** architecture by **alibaba-pai**, optimized for super-resolution detail enhancement. **Model patch.**
60
  - **Qwen-Image-2512-Fun-Controlnet-Union-2602**: Multi-condition ControlNet Union model (5 layer blocks) for the **Qwen-Image-2512** architecture.
61
  - **Qwen-Image-ControlNet-Inpainting**: Dedicated inpainting and editing ControlNet model for the **Qwen-Image** architecture.
62
+ - **Qwen-Image DiffSynth ControlNet Suite** (canny / depth / inpaint): Dedicated single-condition ControlNet models for the **Qwen-Image** architecture developed by **DiffSynth-Studio**. **Model patches.**
 
 
 
63
  - **FLUX.1-dev-ControlNet-Union-Pro-2.0**: Next-generation unified 7-in-1 ControlNet for the **FLUX.1-dev** architecture by **Shakker Labs**.
64
 
65
  ---
66
 
67
  ## 📦 Available Models
68
 
69
+ Sizes are decimal (1 GB = 10\^9 bytes) and exact; the `SHA256` column is the pack's own hash, so a download can be verified against it.
70
+
71
+ | Filename | Base Architecture | Base Model / Author | Supported Conditions | Quantization | File Size | SHA256 | License |
72
+ | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
73
+ | `CN-anytest4_illustrious2_A_hswq_convrot_int8.safetensors` | Illustrious-XL / SDXL | [2vXpSwA7/iroiro-lora (Anytest v4 Variant A)](https://huggingface.co/2vXpSwA7/iroiro-lora) | Multi-condition (Canny, Lineart, Depth, Pose, Structure) | ConvRot INT8 | 1425.1 MB | `be9ff80b2f4895684c7114afff12220394f838179b5c82eadd8ebc46f5d82c25` | Fair AI / OpenRAIL++-M |
74
+ | `CN-anytest4_illustrious2_B_hswq_convrot_int8.safetensors` | Illustrious-XL / SDXL | [2vXpSwA7/iroiro-lora (Anytest v4 Variant B)](https://huggingface.co/2vXpSwA7/iroiro-lora) | Multi-condition (Canny, Lineart, Depth, Pose, Structure) | ConvRot INT8 | 1423.5 MB | `7e34dc3cc3334afa2b9fe5461131a1f61e289a83b1a11413d6c687e94b9dde6e` | Fair AI / OpenRAIL++-M |
75
+ | `controlnet-union-pro-max-sdxl-1.0_hswq_convrot_int8.safetensors` | SDXL 1.0 | [xinsir/controlnet-union-sdxl-1.0](https://huggingface.co/xinsir/controlnet-union-sdxl-1.0) | OpenPose, Depth, Canny, Lineart, Anime Lineart, Scribble, Soft Edge, Normal, Segment, Tile, Inpaint | ConvRot INT8 | 1435.3 MB | `dcdf6f2037205c4a40b1f8863a91827ca3a5b55ec3261e07bb13d2dc855b47fb` | OpenRAIL++-M / Apache-2.0 |
76
+ | `FLUX.1-dev-ControlNet-Union-Pro-2.0_hswq_convrot_int8.safetensors` | FLUX.1-dev | [Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro-2.0](https://huggingface.co/Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro-2.0) | Canny, Depth, Pose, Blur, Gray, Soft Edge, Low Quality | ConvRot INT8 | 2710.6 MB | `fbcba3c76a221fba461c83680f682d3ff7aa813ef554c42e3a5449b981090c46` | Other / Non-Commercial |
77
+ | `Qwen-Image-2512-Fun-Controlnet-Union-2602_hswq_convrot_int8.safetensors` | Qwen-Image-2512 | [alibaba-pai/Qwen-Image-2512-Fun-Controlnet-Union](https://huggingface.co/alibaba-pai/Qwen-Image-2512-Fun-Controlnet-Union) | Canny, HED, Depth, Pose, MLSD, Scribble, Gray, Inpaint | ConvRot INT8 | 1758.7 MB | `5aec39a53f583bade3b7db18f38ed5a01905a0610c2ad2f9f792802178596bc4` | Apache-2.0 |
78
+ | `Qwen-Image-ControlNet-Inpainting_hswq_convrot_int8.safetensors` | Qwen-Image | [alibaba-pai/Qwen-Image-ControlNet-Inpainting](https://huggingface.co/alibaba-pai/Qwen-Image-ControlNet-Inpainting) | Inpainting, Image Editing | ConvRot INT8 | 2120.3 MB | `38f0ab7495029517d4b0bb4187cedcbdfa090a29f0bb50dfa5733f5076a0c451` | Apache-2.0 |
79
+ | `qwen_image_canny_diffsynth_controlnet_hswq_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Canny Edge Detection & Structural Guidance | ConvRot INT8 | 1135.9 MB | `c2899255252b81fd64ef2fae0fbfd86e1496ba04cf10cc95fe45679120f1c996` | Apache-2.0 |
80
+ | `qwen_image_depth_diffsynth_controlnet_hswq_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Geometric Depth Map Guidance | ConvRot INT8 | 1135.9 MB | `7b82a9360169fa99524c03bcf913ce63713930174ad1fc332e553e5646bad44c` | Apache-2.0 |
81
+ | `qwen_image_inpaint_diffsynth_controlnet_hswq_convrot_int8.safetensors` | Qwen-Image | [modelscope/DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) | Masked Inpainting & Localized Editing | ConvRot INT8 | 1135.9 MB | `471b6affc7fcedafb9940fafd6354dd6eb5fac5705e647029c90598c407c82d3` | Apache-2.0 |
82
+ | `Z-Image-Turbo-Fun-Controlnet-Union-2.1-lite-2602-8steps_hswq_convrot_int8.safetensors` | Z-Image-Turbo | [alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1](https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1) | Multi-condition (Canny, Depth, Pose, Color, Scribble, Inpaint) | ConvRot INT8 | 1009.7 MB | `1dd12b2d028b12df6b2a4f93856f8ab161c1edc05125e3b1c987a1694e99dba4` | Apache-2.0 |
83
+ | `Z-Image-Turbo-Fun-Controlnet-Tile-2.1-lite-2601-8steps_hswq_convrot_int8.safetensors` | Z-Image-Turbo | [alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1](https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1) | Tile, Super-Resolution, Detail Enhancing | ConvRot INT8 | 1009.7 MB | `2998f83ef65903b4b6b22fb7ac270ed66f9a2443220563aa43db4188024763a8` | Apache-2.0 |
84
 
85
  ---
86
 
87
+ ## 📐 Measured fidelity
88
 
89
+ Each pack's protection set was chosen by a **measured** aggregate contribution on the model's own output, over 8 seeds, and the acceptance was measured separately on a fixed 256-item set (seed 42) through the same loader the runtime uses. "Gate" is `cos.mean >= 0.99` and `cos.min >= 0.98`.
90
+
91
+ | Pack | the score (mean / min) | the gate | the note |
92
+ | :--- | :--- | :--- | :--- |
93
+ | `CN-anytest4_illustrious2_A` | 0.999626 (forward, relMSE 1.11e-03) | PASS | trajectory-impact selection, 36 protected |
94
+ | `CN-anytest4_illustrious2_B` | 0.992958 / 0.948279 (trajectory) | PASS | trajectory gate, 35 protected |
95
+ | `controlnet-union-pro-max-sdxl-1.0` | 0.999961 | PASS | curve flat, smallest satisfying pack taken (41 protected) |
96
+ | `Qwen-Image-2512-Fun-Controlnet-Union-2602` | 0.999158 / 0.998215 | PASS | floor case: all 77 layers INT8, no protection |
97
+ | `Qwen-Image-ControlNet-Inpainting` | 0.999545 / 0.998887 | PASS | floor case: all 95 layers INT8, no protection |
98
+ | `qwen_image_canny_diffsynth` | 0.9999924905 / 0.9999924096 | PASS | one protected layer (`img_in`), 598x above the runner-up |
99
+ | `qwen_image_depth_diffsynth` | 0.9999924109 / 0.9999923262 | PASS | one protected layer (`img_in`), 376x above the runner-up |
100
+ | `qwen_image_inpaint_diffsynth` | 0.9999921236 / 0.9999919542 | PASS | one protected layer (`img_in`), 147x above the runner-up |
101
+ | `Z-Image-Turbo-Fun-Controlnet-Union` | 0.999862 / 0.999491 (trajectory) | PASS | all 38 layers INT8, no protection |
102
+ | `Z-Image-Turbo-Fun-Controlnet-Tile` | 0.999956 / 0.999813 (trajectory) | PASS | all 38 layers INT8, no protection |
103
+ | `FLUX.1-dev-ControlNet-Union-Pro-2.0` | 0.999771 / 0.999488 | PASS | 12 protected; the all-INT8 state measured 0.999622 / 0.999020 |
104
+
105
+ The per-checkpoint reasoning and the full measured curves are in the HSWQ repository's `md/` guides.
106
 
107
  ---
108
 
109
+ ## 🛠️ Key Features
110
 
111
+ - **All-in-One & Dedicated Condition Control**: Unified multi-condition architectures alongside specialized single-condition models (Canny, Depth, Tile, Inpainting) for fine-grained workflow flexibility.
112
+ - **ConvRot INT8 Precision**: Orthogonal Hadamard rotations redistribute channel-wise outlier spikes across the input dimension before per-channel INT8 quantization, preventing quantization error buildup in deep control layers.
113
+ - **Measured, not assumed, protection**: Where a layer's measured contribution is large it stays FP16 (kept tensors are stored FP16, not FP32). Several packs here need no protection at all; the measured curves say so.
114
+ - **Footprint**: roughly half the unquantized checkpoint, with the INT8 execution path (Tensor Core INT8 matmul) as the additional runtime gain.
115
 
116
+ ---
117
 
118
+ ## 🚀 Usage in ComfyUI
119
 
120
+ Install the loader extension:
121
 
122
  ```bash
123
  cd ComfyUI/custom_nodes
124
  git clone https://github.com/ussoewwin/ComfyUI-HSWQ-Loader-and-Tools.git
125
  ```
126
 
127
+ Then place each file in the directory its type requires (see the table at the top):
128
+
129
+ - **ControlNet** files go in `ComfyUI/models/controlnet/` and load with `ControlNetLoader`.
130
+ - **Model patch** files go in `ComfyUI/models/model_patches/` and load with **`HSWQ Model Patch Loader`** (the stock `ModelPatchLoader` reads the INT8 tensors as plain weights), then apply with `ZImageFunControlnet` / `QwenImageDiffsynthControlnet`.
131
 
132
  ---
133