Update app.py

app.py

@@ -1,4 +1,4 @@
-# app.py — ZeroGPU対応版
+# app.py — ZeroGPU対応 + ポート自動フォールバック
 import gradio as gr
 import spaces
 import torch
@@ -10,14 +10,11 @@ import traceback
 import base64
 import io
 from pathlib import Path
-
-# FastAPI関連（ハイブリッド構成のため維持）
 from fastapi import FastAPI, UploadFile, File, Form, HTTPException
 
 ##############################################################################
 # 0. 設定とヘルパー
 ##############################################################################
-# モデル・LoRA キャッシュを /data に置ける場合はそちらを優先
 PERSIST_BASE = Path("/data")
 CACHE_ROOT   = (PERSIST_BASE / "instantid_cache" if PERSIST_BASE.exists()
                 and os.access(PERSIST_BASE, os.W_OK)
@@ -28,14 +25,15 @@ for d in (MODELS_DIR, LORA_DIR):
     d.mkdir(parents=True, exist_ok=True)
 
 def dl(url: str, dst: Path, attempts: int = 2):
-    """冪等ダウンロード（既存ならスキップ、リトライ付き）"""
-    if dst.exists(): return
+    """冪等ダウンロード（既にあればスキップ、リトライ付き）"""
+    if dst.exists():
+        return
     for i in range(1, attempts + 1):
         print(f"⬇ Downloading {dst.name} (try {i}/{attempts})")
-        if subprocess.call(["wget", "-q", "-O", str(dst), url]) == 0: return
+        if subprocess.call(["wget", "-q", "-O", str(dst), url]) == 0:
+            return
     raise RuntimeError(f"download failed → {url}")
 
-# 1. Asset download (起動時に実行)
 print("— Starting asset download check —")
 BASE_CKPT = MODELS_DIR / "beautiful_realistic_asians_v7_fp16.safetensors"
 dl("https://civitai.com/api/download/models/177164?type=Model&format=SafeTensor&size=pruned&fp=fp16", BASE_CKPT)
@@ -45,58 +43,8 @@ LORA_FILE = LORA_DIR / "ip-adapter-faceid-plusv2_sd15_lora.safetensors"
 dl("https://huggingface.co/h94/IP-Adapter-FaceID/resolve/main/ip-adapter-faceid-plusv2_sd15_lora.safetensors", LORA_FILE)
 print("— Asset download check finished —")
 
-
-# 2. パイプライン初期化関数 (GPU確保後に呼び出される)
-def load_pipeline():
-    from diffusers import (
-        StableDiffusionPipeline, ControlNetModel,
-        DPMSolverMultistepScheduler, AutoencoderKL,
-    )
-    from insightface.app import FaceAnalysis
-
-    print("→ Loading models to GPU …")
-
-    # --- InstantID 主要モデル ---
-    vae      = AutoencoderKL.from_pretrained(
-        "stabilityai/sd-vae-ft-mse",
-        torch_dtype=torch.float16
-    )
-    base     = StableDiffusionPipeline.from_single_file(
-        str(BASE_CKPT),
-        vae=vae,
-        torch_dtype=torch.float16,
-        safety_checker=None,
-        original_config_file="v1-inference.yaml"  # StableDiffusion1.x 互換
-    )
-    control  = ControlNetModel.from_pretrained(
-        "lllyasviel/control_v11p_sd15_openpose",
-        torch_dtype=torch.float16
-    )
-    pipe     = StableDiffusionPipeline(
-        vae=vae,
-        text_encoder=base.text_encoder,
-        tokenizer=base.tokenizer,
-        unet=base.unet,
-        controlnet=control,
-        scheduler=DPMSolverMultistepScheduler.from_config(base.scheduler.config),
-        safety_checker=None,
-        feature_extractor=base.feature_extractor,
-        requires_safety_checker=False
-    ).to("cuda", dtype=torch.float16)
-    pipe.load_lora_weights(str(LORA_FILE))
-    pipe.set_adapters(["ip_adapter_face"], [1.0])
-    pipe.enable_xformers_memory_efficient_attention()
-
-    # --- InsightFace ---
-    face_analyzer = FaceAnalysis(name="antelopev2", providers=["CUDAExecutionProvider"])
-    face_analyzer.prepare(ctx_id=0, det_size=(640, 640))
-
-    print("✓ Model loading complete.")
-    return pipe, face_analyzer
-
-
 ##############################################################################
-# 3. Gradio UI
+# 1. Gradio UI
 ##############################################################################
 with gr.Blocks(title="InstantID × Beautiful Realistic Asians v7") as demo:
     with gr.Row(equal_height=True):
@@ -116,62 +64,49 @@ with gr.Blocks(title="InstantID × Beautiful Realistic Asians v7") as demo:
             btn = gr.Button("生成",variant="primary")
         with gr.Column():
             out_img = gr.Image(label="結果")
-            
-    # .queue() はGradioの通常機能として必要
     demo.queue()
-    
-    def generate_ui(face_img, subj, add, addneg, cfg, ipw, steps, w, h, upscale, up_factor):
-        # 実際の推論関数（省略：ここに InstantID 推論処理を実装）
-        return face_img  # ダミー
 
-    btn.click(
-        fn=generate_ui, 
-        inputs=[face_in,subj_in,add_in,addneg_in,cfg_sld,ip_sld,step_sld,w_sld,h_sld,up_ck,up_fac], 
-        outputs=[out_img]
-    )
+    # ダミー推論（実装は省略）
+    def generate_ui(*args, **kwargs):
+        return Image.new("RGB", (512,768), (127,127,127))
+    btn.click(generate_ui,
+              inputs=[face_in,subj_in,add_in,addneg_in,cfg_sld,ip_sld,step_sld,
+                      w_sld,h_sld,up_ck,up_fac],
+              outputs=[out_img])
 
 ##############################################################################
-# 4. FastAPI エンドポイント（REST API 用）
+# 2. FastAPI ラッパー（REST API）
 ##############################################################################
 app = FastAPI()
 
 @app.post("/api/predict")
-async def predict(
-    face: UploadFile = File(...),
-    subject: str = Form(...),
-    add_prompt: str = Form(""),
-    add_neg: str = Form(""),
-    cfg: float = Form(6.0),
-    ipw: float = Form(0.6),
-    steps: int = Form(20),
-    w: int = Form(512),
-    h: int = Form(768),
-    upscale: bool = Form(True),
-    up_factor: int = Form(2)
-):
+async def predict(face: UploadFile = File(...)):
     try:
-        # 実際の推論ロジック（省略）
-        result_pil_image = Image.open(face.file)  # ダミー
-
+        img = Image.open(face.file)
         buffered = io.BytesIO()
-        result_pil_image.save(buffered, format="PNG")
-        img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
-        
-        return {"image_base64": img_str}
+        img.save(buffered, format="PNG")
+        img_b64 = base64.b64encode(buffered.getvalue()).decode()
+        return {"image_base64": img_b64}
     except Exception as e:
         traceback.print_exc()
         raise HTTPException(status_code=500, detail=str(e))
 
-# GradioアプリをFastAPIアプリにマウント
+# Gradio を FastAPI にマウント
 app = gr.mount_gradio_app(app, demo, path="/")
 
 print("Application startup script finished. Waiting for requests.")
 
-#------------------------------------------------------------------------
-# 5. Uvicorn サーバー起動（Spaces が呼び出すエントリポイント）
-#------------------------------------------------------------------------
+##############################################################################
+# 3. Uvicorn 起動（ポート重複時フォールバック）
+##############################################################################
 if __name__ == "__main__":
-    import uvicorn, os
-    # Hugging Face Spaces が $PORT を渡してくる場合はそれを優先
-    port = int(os.getenv("PORT", 7860))
-    uvicorn.run(app, host="0.0.0.0", port=port, workers=1, log_level="info")
+    import uvicorn
+    port_env = int(os.getenv("PORT", "7860"))
+    try:
+        uvicorn.run(app, host="0.0.0.0", port=port_env, workers=1, log_level="info")
+    except OSError as e:
+        if e.errno == 98 and port_env != 7860:
+            print(f"⚠️ Port {port_env} busy → falling back to 7860")
+            uvicorn.run(app, host="0.0.0.0", port=7860, workers=1, log_level="info")
+        else:
+            raise