Create app.py

app.py

@@ -0,0 +1,91 @@
+import gradio as gr
+import torch
+from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
+
+# Load the base Stable Diffusion model (v1.5) in half-precision for efficiency
+model_id = "runwayml/stable-diffusion-v1-5"
+pipe = StableDiffusionPipeline.from_pretrained(
+    model_id, 
+    torch_dtype=torch.float16, 
+    safety_checker=None   # disable safety checker for faster inference (floor plans are non-NSFW)
+)
+pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+pipe.enable_attention_slicing()  # reduce memory usage, good for running on limited GPU
+
+# Move the pipeline to GPU (if available)
+pipe = pipe.to("cuda" if torch.cuda.is_available() else "cpu")
+
+# Load the FloorPlan LoRA weights into the pipeline
+pipe.load_lora_weights(
+    "ejazhabibdar/sd-FloorPlan-model",  # Hugging Face model hub ID for the LoRA weights
+    weight_name="pytorch_lora_weights.safetensors"  # file containing LoRA weights
+):contentReference[oaicite:1]{index=1}
+
+# (Optional) Enable xFormers for faster/less memory if installed
+# try:
+#     pipe.enable_xformers_memory_efficient_attention()
+# except Exception as e:
+#     pass
+
+# Define a function to generate the floor plan image based on user inputs
+def generate_floorplan(bedrooms, bathrooms, floors, plot_size):
+    # Construct a text prompt from the structured inputs:
+    # We use the LoRA's recommended style: "Floor plan of a [small/big] house, few/many rooms, one/multiple bathrooms, ...":contentReference[oaicite:2]{index=2}
+    prompt_parts = []
+    # Determine house size descriptor
+    if plot_size.lower() in ["small", "medium", "large"]:
+        size_word = "small" if plot_size.lower() == "small" else ("big" if plot_size.lower() == "large" else "")
+        if size_word:
+            prompt_parts.append(f"{size_word} house")
+        else:
+            prompt_parts.append("house")
+    else:
+        prompt_parts.append("house")
+    # Rooms descriptor based on bedrooms count
+    if bedrooms <= 2:
+        prompt_parts.append("few rooms")
+    elif bedrooms >= 5:
+        prompt_parts.append("many rooms")
+    else:
+        prompt_parts.append(f"{bedrooms} rooms")
+    # Bathrooms descriptor
+    if bathrooms == 1:
+        prompt_parts.append("one bathroom")
+    else:
+        prompt_parts.append(f"{bathrooms} bathrooms")
+    # Floors descriptor
+    if floors == 1:
+        # single floor might not need explicit mention
+        prompt_parts.append("one floor")
+    else:
+        prompt_parts.append(f"{floors} floors")
+    # You can add more prompt details (kitchen size, windows) if desired, but we'll keep it simple
+    prompt = "Floor plan of a " + ", ".join(prompt_parts) + "."
+    
+    # Generate the image with the pipeline
+    image = pipe(prompt, 
+                 num_inference_steps=25,   # number of diffusion steps
+                 guidance_scale=7.5        # classifier-free guidance scale
+                 # You can adjust the LoRA strength via cross_attention_kwargs if needed, e.g. {"scale": 0.8}
+                ).images[0]
+    return image
+
+# Build the Gradio interface
+with gr.Blocks() as demo:
+    gr.Markdown("## 🏠 AI Floor Plan Generator\nEnter your desired house parameters and click **Generate** to create a floor plan:")
+    with gr.Row():
+        bedrooms = gr.Slider(label="Bedrooms", minimum=1, maximum=10, value=3, step=1)
+        bathrooms = gr.Slider(label="Bathrooms", minimum=1, maximum=5, value=2, step=1)
+    with gr.Row():
+        floors = gr.Slider(label="Floors", minimum=1, maximum=3, value=1, step=1)
+        plot_size = gr.Dropdown(label="Plot Size", choices=["Small", "Medium", "Large"], value="Medium")
+    generate_btn = gr.Button("Generate Floor Plan")
+    output_image = gr.Image(label="Generated Floor Plan", type="pil")
+    
+    # Connect the input components to the generation function
+    generate_btn.click(fn=generate_floorplan, 
+                       inputs=[bedrooms, bathrooms, floors, plot_size], 
+                       outputs=output_image)
+
+# Launch the app (not strictly required on Spaces, but okay to include)
+demo.launch()