Update app.py

app.py

@@ -8,10 +8,9 @@ import os
 # from diffusers import QwenImageEditInpaintPipeline
 from optimization import optimize_pipeline_
 from diffusers.utils import load_image
-from diffusers import FlowMatchEulerDiscreteScheduler
-from qwenimage.pipeline_qwenimage_edit_inpaint import QwenImageEditInpaintPipeline
-from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
-from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3
+
+from diffusers import QwenImageControlNetModel, QwenImageControlNetInpaintPipeline
+
 import math
 from huggingface_hub import InferenceClient
 
@@ -148,54 +147,27 @@ def use_output_as_input(output_image):
         return gr.update(value=output_image[1])
     return gr.update()
 
-# Initialize Qwen Image Edit pipeline
-# Scheduler configuration for Lightning
-scheduler_config = {
-    "base_image_seq_len": 256,
-    "base_shift": math.log(3),
-    "invert_sigmas": False,
-    "max_image_seq_len": 8192,
-    "max_shift": math.log(3),
-    "num_train_timesteps": 1000,
-    "shift": 1.0,
-    "shift_terminal": None,
-    "stochastic_sampling": False,
-    "time_shift_type": "exponential",
-    "use_beta_sigmas": False,
-    "use_dynamic_shifting": True,
-    "use_exponential_sigmas": False,
-    "use_karras_sigmas": False,
-}
-
-# Initialize scheduler with Lightning config
-scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
-
-
-pipe = QwenImageEditInpaintPipeline.from_pretrained("Qwen/Qwen-Image-Edit", scheduler=scheduler, torch_dtype=torch.bfloat16).to("cuda")
-pipe.load_lora_weights(
-        "lightx2v/Qwen-Image-Lightning", 
-        weight_name="Qwen-Image-Lightning-8steps-V1.1.safetensors"
-    )
-pipe.fuse_lora()
 
-# pipe.transformer.__class__ = QwenImageTransformer2DModel
-pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3())
+base_model = "Qwen/Qwen-Image"
+controlnet_model = "InstantX/Qwen-Image-ControlNet-Inpainting"
 
+controlnet = QwenImageControlNetModel.from_pretrained(controlnet_model, torch_dtype=torch.bfloat16)
 
-# dummy_mask  = load_image("https://github.com/Trgtuan10/Image_storage/blob/main/mask_cat.png?raw=true")
+pipe = QwenImageControlNetInpaintPipeline.from_pretrained(
+    base_model, controlnet=controlnet, torch_dtype=torch.bfloat16
+)
+pipe.to("cuda")
 
-# # --- Ahead-of-time compilation ---
-# optimize_pipeline_(pipe, image=Image.new("RGB", (1328, 1328)), prompt="prompt", mask_image=dummy_mask)
 
 @spaces.GPU(duration=120)
 def infer(edit_images, 
           prompt, 
-          negative_prompt="", 
+          negative_prompt=" ", 
           seed=42, 
           randomize_seed=False, 
           strength=1.0, 
-          num_inference_steps=8, 
-          true_cfg_scale=1.0,
+          num_inference_steps=30, 
+          true_cfg_scale=4.0,
           rewrite_prompt=True,
           progress=gr.Progress(track_tqdm=True)):
     
@@ -213,9 +185,9 @@ def infer(edit_images,
     result_image = pipe(
         prompt=prompt,
         negative_prompt=negative_prompt,
-        image=image,
-        mask_image=mask,
-        strength=strength,
+        control_image=image,
+        control_mask=mask_image,
+        controlnet_conditioning_scale=strength,
         num_inference_steps=num_inference_steps,
         true_cfg_scale=true_cfg_scale,
         generator=torch.Generator(device="cuda").manual_seed(seed)
@@ -244,22 +216,11 @@ css = """
 """
 
 
-with gr.Blocks(css=css) as demo:
-    
-    with gr.Column(elem_id="col-container"):
-        gr.HTML("""
-        <div id="logo-title">
-            <img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/qwen_image_edit_logo.png" alt="Qwen-Image Edit Logo" width="400" style="display: block; margin: 0 auto;">
-            <h2 style="font-style: italic;color: #5b47d1;margin-top: -27px !important;margin-left: 133px;">Inpaint</h2>
-        </div>
-        """)
-        gr.Markdown("""
-        
-        Inpaint images with Qwen Image Edit. [Learn more](https://github.com/QwenLM/Qwen-Image) about the Qwen-Image series. 
-        
-        This demo uses the [Qwen-Image-Lightning](https://huggingface.co/lightx2v/Qwen-Image-Lightning) LoRA with FA3 for accelerated 8-step inference.
-        Try on [Qwen Chat](https://chat.qwen.ai/), or [download model](https://huggingface.co/Qwen/Qwen-Image-Edit) to run locally with ComfyUI or diffusers.
-        """)
+with gr.Blocks(css=css, theme=gr.themes.Citrus()) as demo:
+        gr.HTML("<h1 style='text-align: center'>Qwen-Image with InstantX Inpainting ControlNet</style>")
+        gr.Markdown(
+            "Generate images with the [InstantX/Qwen-Image-ControlNet-Inpainting](https://huggingface.co/InstantX/Qwen-Image-ControlNet-Inpainting) that takes depth, pose and canny conditionings"
+        )
         with gr.Row():
             with gr.Column():
                 edit_image = gr.ImageEditor(
@@ -309,7 +270,7 @@ with gr.Blocks(css=css) as demo:
             
             with gr.Row():
                 strength = gr.Slider(
-                    label="Strength",
+                    label="Conditioning Scale",
                     minimum=0.0,
                     maximum=1.0,
                     step=0.1,
@@ -322,7 +283,7 @@ with gr.Blocks(css=css) as demo:
                     minimum=1.0,
                     maximum=10.0,
                     step=0.5,
-                    value=1.0,
+                    value=4.0,
                     info="Classifier-free guidance scale"
                 )
   
@@ -331,7 +292,7 @@ with gr.Blocks(css=css) as demo:
                     minimum=1,
                     maximum=50,
                     step=1,
-                    value=8,
+                    value=30,
                 )
             rewrite_prompt = gr.Checkbox(
                 label="Enhance prompt (using HF Inference)",