IO_pipeline.py

import os
import json
import base64
import io
from PIL import Image
from huggingface_hub import InferenceClient

# ==========================================
#  RECIPE DIGITALIZER PIPELINE (Image -> JSON)
# ==========================================

class RecipeDigitalizerPipeline:
    def __init__(self):
        print("Connecting to Hugging Face API (Qwen Mode)...")
        self.token = os.getenv("HF_TOKEN")

        # This targets the larger model which uses a DIFFERENT server
        self.model_id = "Qwen/Qwen2.5-VL-72B-Instruct" 

        self.client = InferenceClient(token=self.token)

    def compress_image(self, image_path):
        """
        Resizes the image so it doesn't crash the Free API.
        """
        with Image.open(image_path) as img:
            if img.mode != 'RGB':
                img = img.convert('RGB')

            # Resize: Free API often rejects images larger than 1024x1024
            max_size = 1024
            if max(img.size) > max_size:
                img.thumbnail((max_size, max_size))

            # Save to memory as JPEG
            buffer = io.BytesIO()
            img.save(buffer, format="JPEG", quality=70) # Quality 70 is enough for text

            # Convert to Base64
            encoded_string = base64.b64encode(buffer.getvalue()).decode('utf-8')
            return f"data:image/jpeg;base64,{encoded_string}"

    def run_pipeline(self, image_path):
        prompt = """Extract the recipe from this image.
        Output strictly valid JSON with keys: title, ingredients (list), instructions (list), cuisine_type, difficulty.
        Do not include markdown formatting like ```json, just the raw JSON."""

        try:
            # 1. Compress Image (Solves 400 Bad Request)
            image_url = self.compress_image(image_path)

            # 2. Call Qwen API
            response = self.client.chat.completions.create(
                model=self.model_id,
                messages=[
                    {
                        "role": "user",
                        "content": [
                            {
                                "type": "image_url",
                                "image_url": {"url": image_url}
                            },
                            {"type": "text", "text": prompt}
                        ]
                    }
                ],
                max_tokens=1024
            )

            # 3. Clean Output
            raw_text = response.choices[0].message.content
            # Remove potential markdown fences
            clean_json = raw_text.replace("```json", "").replace("```", "").strip()

            # Extra safety: Find the first { and last }
            start = clean_json.find('{')
            end = clean_json.rfind('}') + 1
            if start != -1 and end != -1:
                clean_json = clean_json[start:end]

            return json.loads(clean_json)

        except Exception as e:
            return {"error": f"Qwen API Error: {str(e)}"}

# --- PART 4: EXECUTION EXAMPLE ---

if __name__ == "__main__":
    
    # 1. AUTHENTICATION FIX (For Colab / Local Env)
    try:
        # If running in Colab, try to fetch from secrets
        try:
            from google.colab import userdata
            hf1_secret = userdata.get('HF_TOKEN')
            os.environ["HF_TOKEN"] = hf1_secret
            print(f"✅ Successfully loaded token from secret HF_TOKEN")
        except ImportError:
            # If running locally, assume env var is already set or manually set here
            pass 

    except Exception as e:
        print(f"⚠️ Warning: Could not load secret 'HF_TOKEN'. Make sure the name in the Key icon is exactly 'HF_TOKEN'.")
        print(f"Error details: {e}")

    # 2. INITIALIZE PIPELINE
    try:
        app = RecipeDigitalizerPipeline()

        # 3. USER INPUT (Update this path to your actual image)
        user_image = "Recipe.jfif" 

        # 4. RUN PIPELINE
        if os.path.exists(user_image):
            print(f"Processing {user_image}...")
            ai_output = app.run_pipeline(user_image)

            # 5. AI OUTPUT
            print("\n--- FINAL DIGITAL OUTPUT ---")
            print(json.dumps(ai_output, indent=4))
        else:
            print(f"❌ Error: Image not found at {user_image}")

    except Exception as e:
        print(f"❌ Application Error: {e}")