Update app.py

app.py

@@ -1,97 +1,79 @@
-from PIL import Image
-from io import BytesIO
-import base64
 import os
+import io
+import base64
+import re
+from PIL import Image
 from groq import Groq
 import gradio as gr
 
-# Read the key name from Hugging Face Environment Variables secrets configuration
-# Ensure "GROQ_API_KEY" is set in your Hugging Face Space settings.
-api_key = os.getenv("GROQ_API_KEY", "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")
-client = Groq(api_key=api_key)
-
-def encode_image(image_path):
-    """Encode the image to base64."""
-    try:
-        # Open the image file
-        image = Image.open(image_path).convert("RGB")
-
-        # Resize the image to a height of 512 while maintaining the aspect ratio
-        base_height = 512
-        h_percent = (base_height / float(image.size[1]))
-        w_size = int((float(image.size[0]) * float(h_percent)))
-        image = image.resize((w_size, base_height), Image.LANCZOS)
-
-        # Convert the image to a byte stream
-        buffered = BytesIO()
-        image.save(buffered, format="JPEG")
-        img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
+# Initialize Groq client
+client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
 
-        return img_str
-    except FileNotFoundError:
-        print(f"Error: The file {image_path} was not found.")
-        return None
-    except Exception as e:
-        print(f"Error: {e}")
+def encode_image(image):
+    if image is None:
         return None
+    buffered = io.BytesIO()
+    image.save(buffered, format="JPEG")
+    return base64.b64encode(buffered.getvalue()).decode('utf-8')
 
-def feifeichat(image):
-    try:
-        if image is None:
-            return "Please upload a photo"
+def feifeichat(input_img):
+    if input_img is None:
+        return '{"error": "No image provided"}'
+    
+    base64_image = encode_image(input_img)
+    
+    prompt = """Analyze the provided image in detail. Generate a structured response adhering strictly to the following criteria:
+1. Provide a concise, clear title summarizing the main subject.
+2. Provide a detailed, engaging description explaining the content, context, key elements, colors, and features visible in the image.
+3. Return the response strictly as valid raw JSON with NO markdown blocks or outer conversational text.
 
-        base64_image = encode_image(image)
-        if not base64_image:
-            return "Error processing image"
+Required JSON format:
+{
+  "title": "Short descriptive title",
+  "description": "Comprehensive detailed description of the image content..."
+}"""
 
-        # Engineered prompt utilizing explicit style rules, forbidden terms, and keyword density targets
-        messages = [
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "text", 
-                        "text": "Analyze the image and return strictly valid JSON format with keys 'title', 'description', 'meta_title' (max 60 chars), 'meta_description' (max 155 chars), 'tags' (array of 5-8 strings), 'style_breakdown' (object with keys: aesthetic, top, bottom, legwear, accessories), and 'faqs' (array of two objects with 'q' and 'a'). Follow these strict copywriting guidelines: 1. Description must be an elegant, descriptive paragraph (3-4 sentences) written like an alternative fashion lookbook. 2. Focus on physical aesthetics: mention terms like 'slender body', 'petite frame', 'youthful appeal', 'cute vibe', and 'sexy style' naturally. 3. Avoid literal camera actions: DO NOT say 'taking a selfie', 'mirror picture', or 'holding a phone'. Describe the pose as casual, relaxed, or playful. 4. Tags must be individual aesthetic search terms (e.g., ['egirl-aesthetic', 'thigh-highs', 'white-tights', 'alternative-fashion'])."
-                    },
-                    {
-                        "type": "image_url",
-                        "image_url": {
-                            "url": f"data:image/jpeg;base64,{base64_image}"
+    try:
+        completion = client.chat.completions.create(
+            model="llama-3.2-11b-vision-preview",
+            messages=[
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": prompt},
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": f"data:image/jpeg;base64,{base64_image}"
+                            }
                         }
-                    }
-                ]
-            }
-        ]
-
-        # Non-streaming call returns complete JSON payload directly to Gradio API
-        response = client.chat.completions.create(
-            model="qwen/qwen3.6-27b",
-            messages=messages,
-            response_format={"type": "json_object"},
-            reasoning_effort="none",
-            temperature=0.7,
-            stream=False
+                    ]
+                }
+            ],
+            temperature=0.2,
+            max_tokens=1024
         )
-
-        return response.choices[0].message.content
-
+        
+        raw_response = completion.choices[0].message.content.strip()
+        
+        # Strip thinking tags if present
+        cleaned_response = re.sub(r'<think>.*?</think>', '', raw_response, flags=re.DOTALL).strip()
+        # Strip Markdown code fencing block if present
+        cleaned_response = re.sub(r'^```(?:json)?\s*', '', cleaned_response)
+        cleaned_response = re.sub(r'\s*```$', '', cleaned_response)
+        
+        return cleaned_response
     except Exception as e:
-        print(f"Error: {e}")
-        return f"An error occurred while processing your request: {e}"
+        return f'{{"error": "{str(e)}"}}'
 
 with gr.Blocks() as demo:
-    gr.Markdown("Image To Flux Prompt")
-    with gr.Tab(label="Image To Flux Prompt"):
-        input_img = gr.Image(label="Input Picture", height=320, type="filepath")
-        output_text = gr.Textbox(label="Flux Prompt")
-        submit_btn = gr.Button(value="Submit")
-
-    # api_name="feifeichat" explicitly exposes /gradio_api/call/feifeichat
-    submit_btn.click(
-        fn=feifeichat,
-        inputs=input_img,
-        outputs=output_text,
-        api_name="feifeichat"
-    )
+    gr.Markdown("# Image Captioning & Analysis API")
+    with gr.Row():
+        input_img = gr.Image(type="pil", label="Input Image")
+        output_text = gr.Textbox(label="JSON Output")
+    
+    submit_btn = gr.Button("Analyze Image")
+    submit_btn.click(fn=feifeichat, inputs=input_img, outputs=output_text, api_name="feifeichat")
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()