From c93073e6f6087df9f8738ccbd3adba16daaeddd5 Mon Sep 17 00:00:00 2001
From: Your Name <you@example.com>
Date: Thu, 11 Dec 2025 18:07:49 +0000
Subject: [PATCH] qwen i2l style

---
 __init__.py    |  3 ++
 nodes.py       | 97 ++++++++++++++++++++++++++++++++++++++++++++++++++
 rh_config.json |  1 +
 3 files changed, 101 insertions(+)
 create mode 100644 __init__.py
 create mode 100644 nodes.py
 create mode 100644 rh_config.json

diff --git a/__init__.py b/__init__.py
new file mode 100644
index 0000000..76028e6
--- /dev/null
+++ b/__init__.py
@@ -0,0 +1,3 @@
+from .nodes import NODE_CLASS_MAPPINGS
+NODE_DISPLAY_NAME_MAPPINGS = {k:k for k,v in NODE_CLASS_MAPPINGS.items()}
+__all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS']
diff --git a/nodes.py b/nodes.py
new file mode 100644
index 0000000..086f00e
--- /dev/null
+++ b/nodes.py
@@ -0,0 +1,97 @@
+import uuid
+import os
+import folder_paths
+from diffsynth.pipelines.qwen_image import (
+    QwenImagePipeline, ModelConfig,
+    QwenImageUnit_Image2LoRAEncode, QwenImageUnit_Image2LoRADecode
+)
+from modelscope import snapshot_download
+from safetensors.torch import save_file
+import torch
+from PIL import Image
+import numpy as np
+
+class RunningHub_ImageQwenI2L_Loader_Style:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+
+            }
+        }
+
+    RETURN_TYPES = ('RH_QwenImageI2LPipeline', )
+    RETURN_NAMES = ('QwenImageI2LPipeline', )
+    FUNCTION = "load"
+    CATEGORY = "RunningHub/ImageQwenI2L"
+
+    def __init__(self):
+        self.vram_config_disk_offload = {
+            "offload_dtype": "disk",
+            "offload_device": "disk",
+            "onload_dtype": "disk",
+            "onload_device": "disk",
+            "preparing_dtype": torch.bfloat16,
+            "preparing_device": "cuda",
+            "computation_dtype": torch.bfloat16,
+            "computation_device": "cuda",
+        }
+        # self.encoder_path = os.path.join(folder_paths.models_dir, 'DiffSynth-Studio', 'General-Image-Encoders')
+        # self.i2l_path = os.path.join(folder_paths.models_dir, 'DiffSynth-Studio', 'Qwen-Image-i2L')
+        # self.processor_path = os.path.join(folder_paths.models_dir, 'DiffSynth-Studio', 'Qwen-Image-Edit')
+
+    def load(self):
+        pipe = QwenImagePipeline.from_pretrained(
+            torch_dtype=torch.bfloat16,
+            device="cuda",
+            model_configs=[
+                ModelConfig(model_id='DiffSynth-Studio/General-Image-Encoders', origin_file_pattern="SigLIP2-G384/model.safetensors", **self.vram_config_disk_offload),
+                ModelConfig(model_id='DiffSynth-Studio/General-Image-Encoders', origin_file_pattern="DINOv3-7B/model.safetensors", **self.vram_config_disk_offload),
+                ModelConfig(model_id='DiffSynth-Studio/Qwen-Image-i2L', origin_file_pattern="Qwen-Image-i2L-Style.safetensors", **self.vram_config_disk_offload),
+            ],
+            processor_config=ModelConfig(model_id='Qwen/Qwen-Image-Edit', origin_file_pattern="processor/"),
+            vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 2,
+        )
+        return (pipe, )
+
+class RunningHub_ImageQwenI2L_LoraGenerator:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "pipeline": ("RH_QwenImageI2LPipeline", ),
+                "training_images": ("IMAGE", ),
+                "seed": ("INT", {"default": 42, "min": 0, "max": 0xffffffffffffffff}),
+            }
+        }
+
+    RETURN_TYPES = ('STRING', 'RH_Lora')
+    RETURN_NAMES = ('lora_name', 'lora')
+    FUNCTION = "generate"
+    CATEGORY = "RunningHub/ImageQwenI2L"
+
+    OUTPUT_NODE = True
+
+    def tensor_2_pil(self, img_tensor):
+        i = 255. * img_tensor.squeeze().cpu().numpy()
+        img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))
+        return img
+
+    def __init__(self):
+        self.lora_name = f"i2l_style_lora_{str(uuid.uuid4())}.safetensors"
+
+    def generate(self, pipeline, training_images, **kwargs):
+        training_images = [self.tensor_2_pil(image) for image in training_images]
+        training_images = [image.convert("RGB") for image in training_images]
+        lora_path = os.path.join(folder_paths.models_dir, 'loras', self.lora_name)
+        with torch.no_grad():
+            embs = QwenImageUnit_Image2LoRAEncode().process(pipeline, image2lora_images=training_images)
+            lora = QwenImageUnit_Image2LoRADecode().process(pipeline, **embs)["lora"]
+        save_file(lora, lora_path)
+        return (self.lora_name, lora_path)
+
+NODE_CLASS_MAPPINGS = {
+    "RunningHub_ImageQwenI2L_Loader(Style)": RunningHub_ImageQwenI2L_Loader_Style,
+    "RunningHub_ImageQwenI2L_LoraGenerator": RunningHub_ImageQwenI2L_LoraGenerator,
+}
+
diff --git a/rh_config.json b/rh_config.json
new file mode 100644
index 0000000..6e0b528
--- /dev/null
+++ b/rh_config.json
@@ -0,0 +1 @@
+{"enable": true, "untracked_paths": []}