Update app.py

app.py

@@ -1,11 +1,15 @@
 import gradio as gr
+import pandas as pd
 import os
 import json
 import base64
+import requests
 from io import BytesIO
 from PIL import Image
+import torch
 import numpy as np
 from google.generativeai import GenerativeModel, configure
+from urllib.parse import urlparse, parse_qs
 
 # Global variable to store API key
 GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY")
@@ -37,6 +41,15 @@ class NanoBananaImageGenerator:
         img_bytes = img_byte_arr.getvalue()
         return {"inline_data": {"mime_type": "image/png", "data": base64.b64encode(img_bytes).decode('utf-8')}}
 
+    def _load_image_from_url(self, url):
+        try:
+            response = requests.get(url)
+            response.raise_for_status()
+            return Image.open(BytesIO(response.content)).convert("RGB")
+        except Exception as e:
+            print(f"Error loading image from URL {url}: {e}")
+            return None
+
     def build_prompt_for_operation(self, prompt, operation="generate", has_references=False, aspect_ratio="1:1", character_consistency=True):
         aspect_instructions = {
             "1:1": "square format",
@@ -90,8 +103,8 @@ class NanoBananaImageGenerator:
             for i in range(batch_count):
                 try:
                     print(f"Debug: Batch {i+1} - Attempting to call Gemini API.")
-                    # Changed the model to a more stable 'gemini-pro-vision'
-                    response = GenerativeModel("gemini-pro-vision").generate_content(
+                    print(f"Debug: Batch {i+1} - Contents (first part): {content_parts[0]['parts'][0]['text'][:50]}...")
+                    response = GenerativeModel("gemini-2.5-flash-image-preview").generate_content(
                         contents=content_parts,
                         generation_config=generation_config
                     )
@@ -102,11 +115,13 @@ class NanoBananaImageGenerator:
 
                     batch_images = []
                     
+                    # 優先檢查提示詞或候選者是否因安全政策被拒絕
                     if hasattr(response, 'prompt_feedback') and response.prompt_feedback.safety_ratings:
                         operation_log += f"提示詞因安全政策被拒絕: {response.prompt_feedback.safety_ratings}\n"
                         continue
 
                     if not hasattr(response, 'candidates') or not response.candidates:
+                        # 如果沒有候選者，但有其他錯誤資訊，記錄下來
                         operation_log += f"批次 {i+1}: 在回應中未找到任何候選者。完整回應: {str(response)}\n"
                         continue
 
@@ -130,13 +145,11 @@ class NanoBananaImageGenerator:
                     else:
                         operation_log += f"批次 {i+1}: 未找到圖像。請檢視日誌了解詳細資訊。\n"
                 
-                except KeyError as e:
-                    if str(e) == "'Text'":
-                        operation_log += f"批次 {i+1} 錯誤: Gemini API 返回了意外的回應結構。影像生成可能因安全政策、無效輸入或內部 API 問題而失敗。原始錯誤: {type(e).__name__} - {str(e)}\n"
-                    else:
-                        raise e
                 except Exception as e:
-                    operation_log += f"批次 {i+1} 意外錯誤: {type(e).__name__} - {str(e)}\n"
+                    operation_log += f"批次 {i+1} 發生意外錯誤: {type(e).__name__} - {str(e)}\n"
+                    if 'response' in locals() and response is not None:
+                        operation_log += f"除錯: 錯誤時的完整回應 (嘗試轉換為字串): {str(response)}\n"
+                        operation_log += f"除錯: 回應物件類型: {type(response)}\n"
             
             return all_generated_images, operation_log
 
@@ -144,23 +157,23 @@ class NanoBananaImageGenerator:
             operation_log = f"API 呼叫錯誤: {type(e).__name__} - {str(e)}\n"
             return [], operation_log
 
-def generate_image_from_uploads(white_background_img, reference_img, prompt):
+def generate_image(white_background_url, reference_image_url, prompt):
     image_generator = NanoBananaImageGenerator(api_key=GEMINI_API_KEY)
     
     if not GEMINI_API_KEY:
         return None, "錯誤: GEMINI_API_KEY 環境變數未設定。"
     
     encoded_images = []
-    
-    # Correctly check for valid uploaded images
-    if white_background_img is not None and isinstance(white_background_img, np.ndarray):
-        encoded_images.append(image_generator._image_to_base64(Image.fromarray(white_background_img).convert("RGB")))
+    wb_image = image_generator._load_image_from_url(white_background_url)
+    if wb_image:
+        encoded_images.append(image_generator._image_to_base64(wb_image))
 
-    if reference_img is not None and isinstance(reference_img, np.ndarray):
-        encoded_images.append(image_generator._image_to_base64(Image.fromarray(reference_img).convert("RGB")))
+    ref_image = image_generator._load_image_from_url(reference_image_url)
+    if ref_image:
+        encoded_images.append(image_generator._image_to_base64(ref_image))
     
     if not encoded_images:
-        return None, "錯誤: 請上傳至少一張圖片。"
+        return None, "錯誤: 無法從提供的 URL 載入任何圖片。"
 
     has_references = len(encoded_images) > 0
     final_prompt = image_generator.build_prompt_for_operation(
@@ -181,25 +194,92 @@ def generate_image_from_uploads(white_background_img, reference_img, prompt):
     else:
         return None, operation_log
 
+def read_google_sheet(sheet_url):
+    try:
+        def build_csv_url(url: str) -> str:
+            parsed = urlparse(url)
+            path_parts = parsed.path.strip("/").split("/")
+            doc_id = None
+            if len(path_parts) >= 3 and path_parts[0] == "spreadsheets" and path_parts[1] == "d":
+                doc_id = path_parts[2]
+            qs_gid = parse_qs(parsed.query).get("gid", [None])[0]
+            frag_gid = None
+            if parsed.fragment:
+                frag_qs = parse_qs(parsed.fragment)
+                frag_gid = frag_qs.get("gid", [None])[0]
+            gid = qs_gid or frag_gid or "0"
+            if doc_id:
+                return f"https://docs.google.com/spreadsheets/d/{doc_id}/export?format=csv&gid={gid}"
+            if "/export" in parsed.path and "format=csv" in parsed.query:
+                return url
+            return url.replace("/edit#gid=0", "/export?format=csv&gid=0")
+        csv_url = build_csv_url(sheet_url)
+        print(f"Attempting to read CSV from: {csv_url}")
+        df = pd.read_csv(csv_url, engine='python', on_bad_lines='warn')
+        print("Successfully read Google Sheet.")
+        return df
+    except Exception as e:
+        print(f"Error reading Google Sheet: {e}")
+        raise gr.Error(f"Error reading Google Sheet: {e}")
+
+def process_sheet_data(sheet_url):
+    try:
+        df = read_google_sheet(sheet_url)
+        if df.shape[1] < 3:
+            error_msg = f"Error: Google Sheet has only {df.shape[1]} columns, but 3 are expected (White Background URL, Reference Image URL, Prompt)."
+            print(error_msg)
+            raise gr.Error(error_msg)
+        white_background_urls = df.iloc[:, 0].tolist()
+        reference_image_urls = df.iloc[:, 1].tolist()
+        prompts = df.iloc[:, 2].tolist()
+        data = []
+        for i, (wb, ref, p) in enumerate(zip(white_background_urls, reference_image_urls, prompts)):
+            if pd.notna(wb) and pd.notna(ref) and pd.notna(p):
+                data.append([i, wb, ref, p])
+        print(f"Processed {len(data)} valid rows.")
+        return data
+    except Exception as e:
+        print(f"Error processing sheet data: {e}")
+        raise gr.Error(f"Error processing sheet data: {e}")
+
+def generate_image_for_row(row_index, dataframe_data):
+    if not isinstance(dataframe_data, pd.DataFrame) or not (0 <= row_index < len(dataframe_data)):
+        return None, "Error: Invalid row index or dataframe data not loaded."
+    row = dataframe_data.iloc[row_index]
+    white_background_url = row.iloc[1]
+    reference_image_url = row.iloc[2]
+    prompt = row.iloc[3]
+    return generate_image(white_background_url, reference_image_url, prompt)
+
 if __name__ == "__main__":
     with gr.Blocks() as demo:
         gr.Markdown("# AutoLS Gradio Image Generator")
-        gr.Markdown("請上傳圖片並輸入提示詞，進行影像生成。")
-        
+        gr.Markdown("Enter the Google Sheet URL to process image generation requests.")
+        sheet_url_input = gr.Textbox(label="Google Sheet URL", value="https://docs.google.com/spreadsheets/d/1G3olHxydDIbnyXdh5nnw5TG0akZFeMeYm-25JmCGDLg/edit?gid=0#gid=0")
+        process_button = gr.Button("Process Sheet")
+        processed_df_state = gr.State()
+        output_dataframe = gr.DataFrame(
+            headers=["Index", "白背圖URL", "參考圖URL", "提示詞"],
+            col_count=(4, "fixed"),
+            interactive=False
+        )
         with gr.Row():
-            white_background_input = gr.Image(type="numpy", label="上傳白背圖", show_label=True)
-            reference_image_input = gr.Image(type="numpy", label="上傳參考圖", show_label=True)
-            
-        prompt_input = gr.Textbox(label="提示詞", placeholder="例如：加上一個舒適的木製椅子。")
-        generate_button = gr.Button("生成圖片")
-
-        generated_image_output = gr.Image(label="生成的圖片", show_label=True)
-        operation_log_output = gr.Textbox(label="操作日誌", lines=5, show_label=True)
-
-        generate_button.click(
-            fn=generate_image_from_uploads,
-            inputs=[white_background_input, reference_image_input, prompt_input],
+            row_index_input = gr.Number(label="Row Index to Generate", precision=0, value=0)
+            generate_selected_button = gr.Button("Generate Image for Selected Row")
+        generated_image_output = gr.Image(label="Generated Image")
+        operation_log_output = gr.Textbox(label="Operation Log", lines=5)
+        process_button.click(
+            fn=process_sheet_data,
+            inputs=sheet_url_input,
+            outputs=output_dataframe
+        ).success(
+            fn=lambda x: x,
+            inputs=output_dataframe,
+            outputs=processed_df_state
+        )
+        generate_selected_button.click(
+            fn=generate_image_for_row,
+            inputs=[row_index_input, processed_df_state],
             outputs=[generated_image_output, operation_log_output]
         )
-
     demo.launch()