Upload 2 files

app.py

@@ -7,6 +7,7 @@ import jwt
 import io
 from PIL import Image
 
+# Пытаемся импортировать Google GenAI
 try:
     from google import genai
     from google.genai import types
@@ -16,26 +17,65 @@ except ImportError as e:
     HAS_GENAI = False
     GENAI_ERR = str(e)
 
+# Пытаемся импортировать OpenAI
+try:
+    from openai import OpenAI
+    HAS_OPENAI = True
+except ImportError:
+    HAS_OPENAI = False
+
 # --- SECRETS AND CONFIGURATION ---
 GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY") 
 KLING_ACCESS_KEY = os.getenv("KLING_ACCESS_KEY")
 KLING_SECRET_KEY = os.getenv("KLING_SECRET_KEY")
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") # НОВЫЙ СЕКРЕТ ДЛЯ GPT-4 Vision
 
-# --- ОБНОВЛЕННЫЙ ПРОМПТ ДЛЯ ФОТО (Banana/GenAI) ---
+# --- STATIC PROMPT FOR PHOTO ---
 SYSTEM_PROMPT = "Hyper-realistic Instagram influencer lifestyle photo, shot on iPhone 15 Pro. The exact person [refer to subject images] is looking directly at the camera with a warm, highly attractive, and confident smile showing natural bright white teeth. Healthy, clear, highly realistic skin texture (natural, not overly airbrushed, but free of blemishes or rough spots, with a soft flattering glow). They are casually holding the product [refer to product images] close to the camera, showing it to the viewers. Flattering soft daylight from a window combined with a subtle ambient glow on the face. The background is a stylish, slightly eclectic modern loft with unique decor, lush indoor plants, and rich textures, giving a lived-in, creative, and non-generic vibe. Authentic social media blogger aesthetic, exact facial likeness, highly detailed, and visually pleasing."
 
 def generate_kling_token(ak, sk):
-    headers = {
-        "alg": "HS256",
-        "typ": "JWT"
-    }
-    payload = {
-        "iss": ak,
-        "exp": int(time.time()) + 1800,
-        "nbf": int(time.time()) - 5
-    }
+    headers = {"alg": "HS256", "typ": "JWT"}
+    payload = {"iss": ak, "exp": int(time.time()) + 1800, "nbf": int(time.time()) - 5}
     return jwt.encode(payload, sk, algorithm="HS256", headers=headers)
 
+# Функция для вызова GPT-4o Vision
+def generate_video_prompt_with_gpt(base64_image_for_gpt):
+    if not HAS_OPENAI or not OPENAI_API_KEY:
+        # Fallback, если ключа нет
+        return "Natural handheld UGC-style video. The influencer smiles warmly, showing the product to the camera. Subtle micro-shakes, photorealistic, cinematic lighting, casual authentic vibe."
+    
+    try:
+        client = OpenAI(api_key=OPENAI_API_KEY)
+        response = client.chat.completions.create(
+            model="gpt-4o",
+            messages=[
+                {
+                    "role": "system",
+                    "content": "You are an expert AI video generation prompter. Your goal is to write a short, highly detailed prompt for Kling AI (Image-to-Video model). Output ONLY the prompt text, nothing else."
+                },
+                {
+                    "role": "user",
+                    "content":[
+                        {
+                            "type": "text", 
+                            "text": "Analyze this starting frame. Write a prompt to animate it. Rules:\n1. Keep it under 400 characters.\n2. Emphasize 'UGC-style', 'handheld smartphone camera', 'subtle micro-shakes'.\n3. Describe natural movements (e.g., 'subtle head tilt', 'natural blinking', 'casually showing the product').\n4. Maintain a photorealistic and authentic vibe."
+                        },
+                        {
+                            "type": "image_url", 
+                            "image_url": {"url": f"data:image/jpeg;base64,{base64_image_for_gpt}"}
+                        }
+                    ]
+                }
+            ],
+            max_tokens=150,
+            temperature=0.7
+        )
+        return response.choices[0].message.content.strip()
+    except Exception as e:
+        print(f"GPT Error: {str(e)}")
+        return "Natural handheld UGC-style video, subtle micro-shakes, smiling and showing product, highly realistic."
+
+
 def create_photo(blogger_images, product_images):
     logs = ""
     if not blogger_images or not product_images:
@@ -65,7 +105,7 @@ def create_photo(blogger_images, product_images):
         yield logs, None
         
         client = genai.Client(api_key=GOOGLE_API_KEY)
-        contents = [SYSTEM_PROMPT]
+        contents =[SYSTEM_PROMPT]
         
         for img in blogger_images:
             path = img if isinstance(img, str) else img.name
@@ -112,53 +152,51 @@ def create_photo(blogger_images, product_images):
 
 def create_video(generated_photo_path):
     logs = ""
-    # Защита от случайного нажатия до генерации фото
+    gpt_prompt = ""
+    
     if not generated_photo_path:
         raise gr.Error("Please click 'Create Photo' first to generate an image for the video!")
 
-    logs += "[System] Initializing Kling AI video generation...\n"
-    yield logs, None
-    time.sleep(1)
+    logs += "[System] Initializing Video Pipeline...\n"
+    # Возвращаем 3 значения: логи, промпт (пока пустой), видео (None)
+    yield logs, gpt_prompt, None
+    time.sleep(0.5)
 
-    # DEMO FALLBACK
+    # --- 1. ПОДГОТОВКА ИЗОБРАЖЕНИЯ ---
+    logs += "[Storage] Preparing base image for analysis and animation...\n"
+    yield logs, gpt_prompt, None
+    
+    img_obj = Image.open(generated_photo_path)
+    if img_obj.mode != 'RGB':
+        img_obj = img_obj.convert('RGB')
+        
+    buffered = io.BytesIO()
+    img_obj.save(buffered, format="JPEG", quality=95)
+    base64_image_clean = base64.b64encode(buffered.getvalue()).decode('utf-8')
+    
+    # --- 2. ВЫЗОВ GPT-4o VISION ---
+    logs += "[GPT-4o Vision] Analyzing generated photo to write perfect Kling prompt...\n"
+    yield logs, gpt_prompt, None
+    
+    # Отправляем в GPT (GPT нужно передать чистый base64, функция сама добавит префикс)
+    gpt_prompt = generate_video_prompt_with_gpt(base64_image_clean)
+    
+    logs += "[GPT-4o Vision] Prompt generated successfully! Sending to Kling AI...\n"
+    # Обновляем UI: теперь в текстовом окне появится текст от GPT!
+    yield logs, gpt_prompt, None
+
+    # --- 3. ВЫЗОВ KLING API ---
     if not KLING_ACCESS_KEY or not KLING_SECRET_KEY:
         logs += "[Warning] Kling API Keys are not set! Running in Demo Fallback mode...\n"
-        yield logs, None
-        time.sleep(1)
-        logs += "[Kling API] POST /v1/videos/image2video -> Sending data...\n"
-        yield logs, None
-        time.sleep(1.5)
-        logs += "[Kling API] Task ID: kl_demo_98765. Status: SUBMITTED...\n"
-        yield logs, None
+        yield logs, gpt_prompt, None
         time.sleep(2)
-        logs += "[Kling API] Status: PROCESSING...\n"
-        yield logs, None
-        time.sleep(2.5)
         logs += "[Success] Status: SUCCEED. Video is ready!\n"
-        yield logs, "https://www.w3schools.com/html/mov_bbb.mp4"
+        yield logs, gpt_prompt, "https://www.w3schools.com/html/mov_bbb.mp4"
         return
 
-    # REAL API CALL
     try:
-        logs += "[Security] Generating 30-min JWT Token using Access & Secret Keys...\n"
-        yield logs, None
         api_token = generate_kling_token(KLING_ACCESS_KEY, KLING_SECRET_KEY)
         
-        # --- БЕРЕМ СГЕНЕРИРОВАННОЕ ФОТО ИЗ UI ---
-        logs += "[Storage] Loading generated photo from UI...\n"
-        yield logs, None
-        
-        img_obj = Image.open(generated_photo_path)
-        if img_obj.mode != 'RGB':
-            img_obj = img_obj.convert('RGB')
-            
-        buffered = io.BytesIO()
-        img_obj.save(buffered, format="JPEG", quality=95)
-        base64_image = base64.b64encode(buffered.getvalue()).decode('utf-8')
-        
-        logs += "[Storage] Generated image prepared for Kling AI.\n"
-        yield logs, None
-        
         headers = {
             "Authorization": f"Bearer {api_token}",
             "Content-Type": "application/json"
@@ -166,16 +204,16 @@ def create_video(generated_photo_path):
         
         payload = {
             "model_name": "kling-v2-6",
-            "image": base64_image,
-            "prompt": "Natural handheld UGC-style video of a young man holding a small WD-40 spray can in his hand, speaking casually to camera like filming a story at home. The camera is front-facing smartphone style, slight natural hand movement, subtle micro-shakes. He slightly tilts the can toward the camera to show the label, rotates it a bit, then looks back at the lens and smiles while talking. Soft daylight from window, cozy apartment background, shallow depth of field. Realistic facial expressions, natural blinking, small head movements, authentic body language. The product remains clearly visible in frame. Casual authentic vibe, not overacted, not commercial studio style.",
-            "negative_prompt": "blur, distort, low quality, bad anatomy",
+            "image": base64_image_clean, # Kling требует чистый Base64
+            "prompt": gpt_prompt,        # Передаем промпт, сгенерированный GPT!
+            "negative_prompt": "blur, distort, low quality, bad anatomy, artificial, overacted",
             "duration": "5",
             "mode": "pro",
             "sound": "off"
         }
 
         logs += f"[Kling API] POST /v1/videos/image2video -> Starting task...\n"
-        yield logs, None
+        yield logs, gpt_prompt, None
         
         post_url = "https://api-singapore.klingai.com/v1/videos/image2video"
         response = requests.post(post_url, headers=headers, json=payload)
@@ -184,12 +222,12 @@ def create_video(generated_photo_path):
         if response.status_code != 200 or resp_json.get("code") != 0:
             error_msg = resp_json.get('message', 'Unknown error')
             logs += f"[Error] API Creation Failed: {error_msg}\n"
-            yield logs, None
+            yield logs, gpt_prompt, None
             return
             
         task_id = resp_json.get("data", {}).get("task_id")
         logs += f"[Kling API] Task ID: {task_id}. Status: SUBMITTED...\n"
-        yield logs, None
+        yield logs, gpt_prompt, None
         
         get_url = f"https://api-singapore.klingai.com/v1/videos/image2video/{task_id}"
         
@@ -200,7 +238,7 @@ def create_video(generated_photo_path):
             
             if poll_json.get("code") != 0:
                 logs += f"[Error] API Polling Failed: {poll_json.get('message')}\n"
-                yield logs, None
+                yield logs, gpt_prompt, None
                 break
                 
             data = poll_json.get("data", {})
@@ -208,73 +246,94 @@ def create_video(generated_photo_path):
             
             if task_status == "succeed":
                 task_result = data.get("task_result", {})
-                videos = task_result.get("videos", [])
+                videos = task_result.get("videos",[])
                 
                 if videos and "url" in videos[0]:
                     video_url = videos[0]["url"]
                     logs += "[Success] Status: SUCCEED. Video generation finished!\n"
-                    yield logs, video_url
+                    yield logs, gpt_prompt, video_url
                 else:
                     logs += "[Error] Status SUCCEED, but no video URL found.\n"
-                    yield logs, None
+                    yield logs, gpt_prompt, None
                 break
                 
             elif task_status == "failed":
                 fail_msg = data.get("task_status_msg", "Unknown reason")
                 logs += f"[Error] Task FAILED. Reason: {fail_msg}\n"
-                yield logs, None
+                yield logs, gpt_prompt, None
                 break
                 
             else:
                 logs += f"[Kling API] Status: {task_status.upper()}... Waiting 5s...\n"
-                yield logs, None
+                yield logs, gpt_prompt, None
                 
     except Exception as e:
         logs += f"[Exception] Internal error occurred: {str(e)}\n"
-        yield logs, None
+        yield logs, gpt_prompt, None
 
 
 # --- CUSTOM UI STYLES ---
 custom_css = (
-    ".container { max-width: 1100px; margin: auto; }\n"
+    ".container { max-width: 1200px; margin: auto; }\n"
     ".output-media { border-radius: 16px; box-shadow: 0 10px 25px rgba(0,0,0,0.1); }\n"
     "button.primary { background: linear-gradient(90deg, #6366f1, #a855f7); border: none; }\n"
     "button.primary:hover { transform: translateY(-2px); box-shadow: 0 5px 15px rgba(168, 85, 247, 0.4); transition: all 0.3s ease; }\n"
-    ".log-window textarea { background-color: #1e1e1e !important; color: #4ade80 !important; font-family: 'Courier New', Courier, monospace !important; font-size: 13px !important; }"
+    ".log-window textarea { background-color: #1e1e1e !important; color: #4ade80 !important; font-family: 'Courier New', Courier, monospace !important; font-size: 13px !important; }\n"
+    ".prompt-box textarea { font-size: 14px !important; color: #374151 !important; }"
 )
 
 with gr.Blocks() as demo:
-    gr.Markdown("<h1 style='text-align: center;'>✨ AI Content Creator Pro</h1>")
-    gr.Markdown("<p style='text-align: center; color: gray;'>Hyper-realistic UGC Generation: Gemini 3 Pro (Photo) & Kling AI (Video)</p>")
+    gr.Markdown("<h1 style='text-align: center;'>✨ AI UGC Content Pipeline</h1>")
+    gr.Markdown("<p style='text-align: center; color: gray;'>GenAI (Photo) ➡️ GPT-4o Vision (Analysis) ➡️ Kling AI (Video)</p>")
     
     with gr.Row():
+        # ЛЕВАЯ КОЛОНКА
         with gr.Column(scale=1):
-            gr.Markdown("### 📥 Inputs")
-            
+            gr.Markdown("### 📥 1. Image Inputs")
             with gr.Group():
                 blogger_imgs = gr.File(label="📸 Blogger (1-5 photos)", file_count="multiple", file_types=["image"])
                 product_imgs = gr.File(label="🛍️ Product (1-5 photos)", file_count="multiple", file_types=["image"])
             
+            gr.Markdown("### 🧠 2. AI Prompts")
+            # Статичное окно для фото
+            banana_prompt_ui = gr.Textbox(
+                label="GenAI Photo Prompt (Static)", 
+                value=SYSTEM_PROMPT, 
+                interactive=False, 
+                lines=4,
+                elem_classes="prompt-box"
+            )
+            
+            # Динамическое окно для видео (будет заполняться GPT)
+            kling_prompt_ui = gr.Textbox(
+                label="Kling Video Prompt (Auto-generated by GPT-4o Vision)", 
+                placeholder="Click 'Create Video'. GPT-4o will analyze the generated photo and write the perfect prompt here...", 
+                interactive=True, 
+                lines=4,
+                elem_classes="prompt-box"
+            )
+
             with gr.Row():
-                btn_photo = gr.Button("Create Photo", variant="primary", size="lg")
-                btn_video = gr.Button("Create Video", variant="primary", size="lg")
+                btn_photo = gr.Button("Step 1: Create Photo", variant="primary", size="lg")
+                btn_video = gr.Button("Step 2: Create Video", variant="primary", size="lg")
                 
             logs_output = gr.Textbox(
                 label="🖥️ System Logs", 
-                lines=8, 
+                lines=6, 
                 interactive=False, 
                 elem_classes="log-window",
                 placeholder="Generation logs will appear here in real-time..."
             )
                 
+        # ПРАВАЯ КОЛОНКА
         with gr.Column(scale=1):
-            gr.Markdown("### 📤 Output")
+            gr.Markdown("### 📤 3. Outputs")
             
             with gr.Tabs():
-                with gr.Tab("Photo"):
-                    out_photo = gr.Image(label="Photo Result", elem_classes="output-media", type="filepath")
-                with gr.Tab("Video"):
-                    out_video = gr.Video(label="Video Result", elem_classes="output-media")
+                with gr.Tab("Photo Result"):
+                    out_photo = gr.Image(label="Generated Photo", elem_classes="output-media", type="filepath")
+                with gr.Tab("Video Result"):
+                    out_video = gr.Video(label="Animated UGC Video", elem_classes="output-media")
             
     btn_photo.click(
         fn=create_photo, 
@@ -282,10 +341,11 @@ with gr.Blocks() as demo:
         outputs=[logs_output, out_photo]
     )
     
+    # Теперь возвращаем 3 значения: логи, промпт в текстовое окно, и видео
     btn_video.click(
         fn=create_video, 
         inputs=[out_photo], 
-        outputs=[logs_output, out_video]
+        outputs=[logs_output, kling_prompt_ui, out_video]
     )
 
 if __name__ == "__main__":

requirements.txt

@@ -2,4 +2,5 @@ gradio>=4.0.0
 requests
 PyJWT
 google-genai
-pillow
+pillow
+openai