Update app.py

app.py

@@ -9,76 +9,58 @@ import io
 
 app = FastAPI()
 
-# =========================
-# Load base model (CPU)
-# =========================
-print("Loading base model...")
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float32,
-    safety_checker=None
+MODEL_REPO = "ebraam1/interior-sd-models"
+MODEL_FILE = "Interior.safetensors"
+
+print("Downloading model file...")
+model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
+
+print("Loading model on CPU (this may take a while)...")
+pipe = StableDiffusionPipeline.from_single_file(
+    model_path,
+    torch_dtype=torch.float32,          # ضروري لـ CPU
+    safety_checker=None,                # تسريع مع تجاهل فلتر NSFW
+    requires_safety_checker=False
 ).to("cpu")
 
+# تحسينات لتقليل استهلاك الميموري على الـ CPU
 pipe.enable_attention_slicing()
 pipe.enable_vae_slicing()
-
-# =========================
-# FORCE enable PEFT backend
-# =========================
-import peft  # 🔥 مهم جدًا
-
-# =========================
-# Load LoRA
-# =========================
-LORA_PATH = hf_hub_download(
-    repo_id="ebraam1/interior-sd-models",
-    filename="Interior_lora.safetensors"
-)
-
-print("Loading LoRA...")
-
-pipe.load_lora_weights(LORA_PATH)
-pipe.fuse_lora(lora_scale=0.8)
+pipe.enable_sequential_cpu_offload()    # ينقل أجزاء الموديل للـ CPU حسب الحاجة
 
 print("Model ready 🔥")
 
-# =========================
 class Prompt(BaseModel):
     prompt: str
 
-def to_bytes(img):
+def to_bytes(img: Image.Image):
     buf = io.BytesIO()
     img.save(buf, format="PNG")
     buf.seek(0)
     return buf
 
-# =========================
 @app.post("/txt2img")
 def generate(data: Prompt):
-
     image = pipe(
         data.prompt,
-        num_inference_steps=6,
-        guidance_scale=5,
-        height=256,
-        width=256
+        num_inference_steps=10,         # ممكن تقلل لـ 6 لو عايز سرعة
+        guidance_scale=7.5,
+        height=512,
+        width=512
     ).images[0]
 
     return StreamingResponse(to_bytes(image), media_type="image/png")
 
-
 @app.post("/img2img")
 async def img2img_api(file: UploadFile = File(...), prompt: str = ""):
-
-    img = Image.open(io.BytesIO(await file.read())).convert("RGB").resize((256,256))
+    img = Image.open(io.BytesIO(await file.read())).convert("RGB").resize((512, 512))
 
     image = pipe(
         prompt=prompt,
         image=img,
         strength=0.6,
-        num_inference_steps=6,
-        guidance_scale=5
+        num_inference_steps=10,
+        guidance_scale=7.5
     ).images[0]
 
     return StreamingResponse(to_bytes(image), media_type="image/png")