app.py

from fastapi import FastAPI, Request
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from diffusers import StableDiffusionPipeline
from io import BytesIO
from PIL import Image
import base64
import torch

app = FastAPI(
    title="Floor Plan Generator API",
    description="Generates clean architectural floor plans from text prompts using maria26/Floor_Plan_LoRA on Stable Diffusion.",
    version="1.0"
)

# Add CORS middleware (allows your Vercel frontend to call the API)
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Allow all origins for public demo (restrict to your Vercel URL in production)
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

@app.get("/")
def root():
    return {"message": "Floor Plan API live 🚀", "docs": "/docs"}

# Device setup (works on CPU or GPU)
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float16 if device == "cuda" else torch.float32

# Load base Stable Diffusion v1.5 + LoRA
print("Loading Stable Diffusion model...")
generator = StableDiffusionPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    torch_dtype=dtype,
    safety_checker=None,  # No NSFW risk for floor plans
).to(device)

# Apply the Floor Plan LoRA
generator.load_lora_weights("maria26/Floor_Plan_LoRA")

# Optimizations for speed and memory (works great on CPU too)
generator.enable_attention_slicing()
if device == "cuda":
    generator.enable_model_cpu_offload()  # Helps on limited GPU VRAM

print("Model loaded successfully!")

@app.post("/generate")
async def generate(request: Request):
    try:
        data = await request.json()
        text = data.get("prompt")
        if not text or not text.strip():
            return JSONResponse({"error": "prompt is required and cannot be empty"}, status_code=400)

        # Strong negative prompt for clean black-and-white architectural plans
        negative_prompt = "blurry, low quality, text, letters, numbers, watermark, people, furniture, colored, photorealistic, 3d render"

        image = generator(
            prompt=text,
            negative_prompt=negative_prompt,
            num_inference_steps=30,      # Good balance (increase to 50 for better quality on CPU)
            guidance_scale=9.0,         # Strong prompt adherence
            height=512,
            width=512,
        ).images[0]

        buffer = BytesIO()
        image.save(buffer, format="PNG")
        buffer.seek(0)
        encoded = base64.b64encode(buffer.getvalue()).decode("utf-8")

        return {"image": encoded}

    except Exception as e:
        print(f"Generation error: {str(e)}")
        return JSONResponse({"error": str(e)}, status_code=500)