Upload offline_update.py

offline_update.py

@@ -0,0 +1,522 @@
+from openai import OpenAI
+import urllib
+import requests
+import base64
+import os
+import ast
+import cv2
+from io import BytesIO
+from PIL import Image
+from tempfile import NamedTemporaryFile
+import time
+from zipfile import ZipFile
+import gradio as gr
+from docx import Document
+import numpy as np
+
+
+api_key = os.environ['OPENAI_API_KEY']
+brandfolder_api = os.environ['BRANDFOLDER_API_KEY']
+
+client_key_dict = {
+    "The Official Moving Company, LLC": 'KXRbpext',
+    "Newmark Commercial Real Estate": 'none',
+    "Test Collection": 'test',
+    'Direct Mail Xperts LLC':'d5J3MdlO'
+}
+
+section_key_dict = {
+    "Original Project Assets": 'c5vm8cnh9jvkjbh7r43qxkv',
+    "Pre-Processed Images": 'rfqf67pbhn8hg6pjcj762q3q',
+}
+
+# Functions
+
+def rename(filename):
+    client = OpenAI()
+    completion = client.chat.completions.create(
+      model="gpt-4o",
+      messages=[
+        {"role": "system", "content": "You are a helpful assistant specializing in renaming files."},
+        {"role": "user", "content": f"Provide a similar name for this filename: {filename}. Only return the filename and use hyphens in the filename."}
+      ]
+    )
+    return completion.choices[0].message.content
+
+
+def get_collection_dict():
+  headers = {
+  'Accept': 'application/json',
+  'Authorization': brandfolder_api
+  }
+
+  r = requests.get('https://brandfolder.com/api/v4/brandfolders/988cgqcg8xsrr5g9h7gtsqkg/collections?per=300', params={
+      # use a dict with your desired URL parameters here
+  }, headers=headers)
+
+  temp = r.json()['data']
+
+  collection_dict = {item['attributes']['name']:item['id'] for item in temp}
+
+  return collection_dict
+
+
+def get_collection_names():
+    collection_dict = get_collection_dict()
+    return list(collection_dict.keys())
+
+
+def get_topical_map_text(path):
+    document = Document(path)
+
+    extracted_text = []
+
+    for paragraph in document.paragraphs:
+        # Get the left indentation of the current paragraph (if any)
+        left_indent = paragraph.paragraph_format.left_indent
+        if left_indent == None:
+          continue
+        else:
+          indent_level = int(left_indent.pt / 20)  # Convert Twips to points and then to a simple indentation level
+
+          # You might want to adjust the logic below depending on how you want to represent indentation
+          indent_symbol = " " * indent_level  # This creates a number of spaces based on the indentation level; adjust as needed
+
+          # Construct the paragraph text with indentation representation
+          formatted_text = f"{indent_symbol}{paragraph.text}"
+          extracted_text.append(formatted_text)
+
+    return "\n".join(extracted_text)
+
+
+def get_asset_info(asset_id):
+  '''
+  Takes information from asset_id
+  Input: asset_id
+  Output: collection_id, collection_name, section_id
+  '''
+  # asset_id = data['data']['attributes']['key']
+  headers = {
+    'Content-Type': 'application/json',
+    'Authorization': brandfolder_api
+  }
+  r = requests.get(f'https://brandfolder.com/api/v4/assets/{asset_id}?include=section,collections,custom_fields,attachments', params={}, headers=headers)
+
+  # gets section_id
+  try:
+    section_id = r.json()['data']['relationships']['section']['data']['id']
+  except:
+    section_id = ''
+
+  # gets collection_id
+  # gets collection_name
+  try:
+    collection_id = r.json()['data']['relationships']['collections']['data'][0]['id']
+    collection_name = [item['attributes']['name'] for item in r.json()['included'] if item['type']=='collections'][0]
+  except:
+    collection_id = ''
+    collection_name = ''
+
+  # gets asset_name, asset_type, and asset_url
+  try:
+    asset_type = [item['attributes']['value'] for item in r.json()['included'] if item['type'] == 'custom_field_values' and item['attributes']['value']=='Photo'][0]
+  except:
+    asset_type = ''
+  try:
+    asset_name = r.json()['data']['attributes']['name']
+  except:
+    asset_name = ''
+  try:
+    access_key = [item['attributes']['value'] for item in r.json()['included'] if item['type'] == 'custom_field_values' and item['attributes']['key'] == 'What is your Access Code?'][0]
+  except:
+    access_key = ''
+  try:
+    asset_url = [item['attributes']['url'] for item in r.json()['included'] if item['type'] == 'attachments'][0]
+  except:
+    asset_url = ''
+  try:
+    client_name = [item['attributes']['value'] for item in r.json()['included'] if item['type'] == 'custom_field_values' and item['attributes']['key'] == 'Client Name'][0]
+  except:
+    client_name = ''
+  try:
+    project_name = [item['attributes']['value'] for item in r.json()['included'] if item['type'] == 'custom_field_values' and item['attributes']['key'] == 'List Project Name Photos Belong To'][0]
+  except:
+    project_name = ''
+
+  return_dict = {
+      "asset_id": asset_id,
+      "section_id": section_id,
+      "collection_id": collection_id,
+      "collection_name": collection_name,
+      "asset_type": asset_type,
+      "asset_name": asset_name,
+      "access_key": access_key,
+      "image_url": asset_url,
+      "client_name": client_name,
+      "project_name": project_name
+  }
+
+  return return_dict
+
+
+def get_seo_tags(image_url, topical_map, new_imgs, attempts=0, max_attempts=10):
+  '''
+  Gets the seo tags and topic/sub-topic classification for an image using OpenAI GPT-4 Vision Preview
+  Input: image path of desired file
+  Output: dict of topic, sub-topic, and seo tags
+  '''
+
+  if attempts > max_attempts:
+    print("Maximum number of retries exceeded.")
+    return {"error": "Max retries exceeded, operation failed."}
+      
+  print('in seo_tags')
+  filenames = ', '.join(new_imgs)
+  
+  # Query for GPT-4
+  topic_map_query = f"""
+  % You are an expert web designer that can only answer questions relevent to the following Topical Map.
+  % Goal: Output the topic, description, caption, seo tags, alt_tags, and filename for this image using the Topical Map provided.
+  
+  % TOPCIAL MAP
+  ```{topical_map}```
+  """
+ # IF YOU CANNOT PROVIDE AN TOPIC FOR EVERY IMAGE AFTER 5 ATTEMPTS, REPLY WITH 'irrelevant'.
+  topic_list = topical_map.split('\n')
+  topic_list = [topic.strip() for topic in topic_list]
+  topic_list.insert(0, "irrelevant")
+  
+  def compress_and_encode_image(url, target_size_mb=20, quality=70):
+      # Fetch the image from the URL
+      response = requests.get(url)
+      response.raise_for_status()  # Ensure the request succeeded
+  
+      # Open the image using Pillow
+      img = Image.open(BytesIO(response.content))
+      img = img.convert('RGB')
+  
+      # Compress the image by adjusting the quality
+      img_bytes = BytesIO()
+      img.save(img_bytes, format='JPEG', quality=quality)
+  
+      # Check if the image size is acceptable
+      while img_bytes.getbuffer().nbytes > (target_size_mb * 1024 * 1024) and quality > 10:
+          quality -= 5
+          img_bytes = BytesIO()
+          img.save(img_bytes, format='JPEG', quality=quality)
+      
+      # Encode the image content to base64
+      encoded_image = base64.b64encode(img_bytes.getvalue()).decode('utf-8')
+  
+      return encoded_image
+
+  base64_image = compress_and_encode_image(image_url)
+
+  # REMOVE WHEN SHARING FILE
+  api_key = os.environ['OPENAI_API_KEY']
+  
+  # Calling gpt-4 vision
+  headers = {
+    "Content-Type": "application/json",
+    "Authorization": f"Bearer {api_key}"
+  }
+# IF YOU CANNOT PROVIDE AN TOPIC FOR EVERY IMAGE AFTER 5 ATTEMPTS, REPLY WITH 'irrelevant'.
+  payload = {
+    "model": "gpt-4o",
+    "response_format": {"type": "json_object"},
+    "messages": [
+      {'role': 'system', 'content': 'You are an expert web designer that can only answer questions relevent to the following topical map.'
+      },
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": topic_map_query +
+                    """
+                    % INSTRUCTIONS
+                        Step 1 - Generate keywords to describe this image
+                        Step 2 - Decide which topic in the Topicla Map this image fall under, using the keywords you generated and the image itself. You are only permitted to use the exact wording of the topic in the topical map. 
+                        Step 2 - Provide a topic-relevant 5 sentence description for the image. Describe the image only using context relevant to the topics in the topical map. 
+                                        Adhere to the following guidelines when crafting your 5 sentence description: 
+                                        - Mention only the contents of the image. 
+                                        - Do not mention the quality of the image. 
+                                        - Ignore all personal information within the image. 
+                                        - Be as specific as possible when identifying tools/items in the image.
+                        Step 3 - Using the description in Step 1, create a 160 character caption. Make sure the caption is less than 160 characters.
+                        Step 4 - Using the description in Step 1, create 3 topic-relevant SEO tags for this image that will drive traffic to our website. The SEO tags must be two words or less. You must give 3 SEO tags.
+                        Step 5 - Using the description in Step 1, provide a topic-relevant SEO alt tag for the image that will enhance how the website is ranked on search engines.
+                        Step 6 - Using the description in Step 1, provide a new and unique filename for the image as well. Use hyphens for the filename. Do not include extension.
+                        Step 7 - YOU ARE ONLY PERMITTED TO OUTPUT THE TOPIC, DESCRIPTION, CAPTION, SEO, ALT_TAG, AND FILENAME IN THE FOLLOWING JSON FORMAT:
+                    
+                    % OUTPUT FORMAT:
+                        {"topic": topic,
+                        "description": description,
+                        "caption": caption,
+                        "seo": [seo],
+                        "alt_tag": [alt tag],
+                        "filename": filename
+                        }
+                    """
+          },
+          {
+            "type": "image_url",
+            "image_url": {
+              "url": f"data:image/jpeg;base64, {base64_image}"
+            }
+          }
+        ]
+      }
+    ],
+    "max_tokens": 300
+  }
+
+  try:
+    response = requests.post("https://api.openai.com/v1/chat/completions", headers=headers, json=payload)
+    response_data = response.json()
+    if response.status_code == 200 and 'choices' in response_data and len(response_data['choices']) > 0:
+        keys = ['topic', 'description', 'caption', 'seo', 'alt_tag', 'filename']
+        json_dict = ast.literal_eval(response.json()['choices'][0]['message']['content'])
+        if json_dict['topic'] not in topic_list:
+            return get_seo_tags(image_url, topical_map, new_imgs, attempts=attempts+1)
+        if set(json_dict.keys()) != set(keys):
+            return get_seo_tags(image_url, topical_map, new_imgs, attempts=attempts+1)
+        
+        
+        return json_dict
+    else:
+        print("API call failed or bad data, retrying...")
+        return get_seo_tags(image_url, topical_map, new_imgs, attempts=attempts + 1)
+  except Exception as e:
+    print("Exception during API call:", str(e))
+    return get_seo_tags(image_url, topical_map, new_imgs, attempts=attempts + 1)
+
+# creates the asset in the client's brand folder
+def create_ai_asset(asset_dict, topical_map, collection_name, new_imgs, tags=True):
+    '''
+    Creates asset from image path. Also creates seo tags, topic, and alt tag for
+    image
+    Input: name of initial asset, name of client, path to image, create tags boolean
+    Output: id of asset
+    '''
+    print(asset_dict)
+    # results from asset_dict
+    topical_map = get_topical_map_text(topical_map)
+    asset_id = asset_dict['asset_id']
+    client_name = asset_dict['client_name']
+    access_key = asset_dict['access_key']
+    client_key = client_key_dict[collection_name]
+    if client_name != collection_name and access_key != client_key:
+      print(f'{collection_name} != {client_name}')
+      print(f'{access_key} != {client_key}')
+      return 
+
+    asset_name = asset_dict['asset_name']
+    
+    collection_id = asset_dict['collection_id']
+    project_name = asset_dict['project_name']
+    if collection_id == '':
+      collection_dict_temp = get_collection_dict()
+      collection_id = collection_dict_temp[client_name]
+    image_url = asset_dict['image_url']
+
+    # get seo, topic, and sub-topic from OpenAI API
+    json_dict = get_seo_tags(image_url, topical_map, new_imgs)
+    if not json_dict:
+        json_dict = get_seo_tags(image_url, topical_map, new_imgs)
+
+    # parsing out results from get_seo_tags
+    topic = json_dict['topic']
+    description = json_dict['description']
+    caption = json_dict['caption']
+    seo_tags = json_dict['seo']
+    alt_tag = json_dict['alt_tag']
+    image_name = json_dict['filename']
+
+    headers = {
+      'Content-Type': 'application/json',
+      'Authorization': brandfolder_api
+    }
+
+    r = requests.get(f'https://brandfolder.com/api/v4/collections/{collection_id}/assets', params={
+        # use a dict with your desired URL parameters here
+    }, headers=headers)
+
+    asset_names = [item['attributes']['name'] for item in r.json()['data']]
+
+    asset_names = new_imgs + asset_names
+    
+    while image_name in asset_names:
+        image_name = rename(image_name)
+        
+
+
+    # posts image with image name
+
+    r = requests.put(f'https://brandfolder.com/api/v4/assets/{asset_id}', json={
+        # use a dict with the POST body here
+        'data': {
+            'attributes': [
+                {
+                    'name': image_name,
+                    'description': description,
+
+                }
+            ]
+        },
+      }, params={}, headers=headers)
+
+
+    # tags and topic payloads
+    tags_payload = {'data': {'attributes': [{'name': tag} for tag in seo_tags]}}
+
+
+    topic_payload = {'data':
+                [
+                    {
+                        'attributes': {
+                            'value': topic
+                            },
+                            'relationships': {
+                                'asset': {
+                                    'data': {'type': 'assets', 'id': asset_id}
+                                            }}
+                    }]}
+
+
+    alt_tag_payload = {'data':
+                    [
+                        {
+                            'attributes': {
+                                'value': alt_tag
+                                },
+                                'relationships': {
+                                    'asset': {
+                                        'data': {'type': 'assets', 'id': asset_id}
+                                                }}
+                        }]}
+
+    year_payload = {'data':
+                    [
+                        {
+                            'attributes': {
+                                'value': 2024
+                                },
+                                'relationships': {
+                                    'asset': {
+                                        'data': {'type': 'assets', 'id': asset_id}
+                                                }}
+                        }]}
+
+    client_payload = {'data':
+                [
+                    {
+                        'attributes': {
+                            'value': client_name
+                            },
+                            'relationships': {
+                                'asset': {
+                                    'data': {'type': 'assets', 'id': asset_id}
+                                            }}
+                    }]}
+    caption_payload = {'data':
+            [
+                {
+                    'attributes': {
+                        'value': caption
+                        },
+                        'relationships': {
+                            'asset': {
+                                'data': {'type': 'assets', 'id': asset_id}
+                                        }}
+                }]}
+
+    project_payload = {'data':
+            [
+                {
+                    'attributes': {
+                        'value': project_name
+                        },
+                        'relationships': {
+                            'asset': {
+                                'data': {'type': 'assets', 'id': asset_id}
+                                        }}
+                }]}
+
+    year_id = 'k8vr5chnkw3nrnrpkh4f9fqm'
+    client_name_id = 'x56t6r9vh9xjmg5whtkmp'
+    # Tone ID: px4jkk2nqrf9h6gp7wwxnhvz
+    # Location ID: nm6xqgcf5j7sw8w994c6sc8h
+    alt_tag_id = 'vk54n6pwnxm27gwrvrzfb'
+    topic_id = '9mcg3rgm5mf72jqrtw2gqm7t'
+    project_name_id = '5zpqwt2r348sjbnc6rpxc96'
+    caption_id = 'cmcbhcc5nmm72v57vrxppw2x'
+    # Original Project Images Section ID: c5vm8cnh9jvkjbh7r43qxkv
+    # Edited Project Images Section ID: 5wpz2s9m3g7ctcjpm4vrt46
+
+    r_asset = requests.post(f'https://brandfolder.com/api/v4/assets/{asset_id}/tags', json=tags_payload, params={}, headers=headers)
+
+    # alt_tags
+    r_topic = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{topic_id}/custom_field_values', json=
+          topic_payload
+        , params={
+      }, headers=headers)
+
+    r_alt_tag = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{alt_tag_id}/custom_field_values', json=
+          alt_tag_payload
+        , params={
+      }, headers=headers)
+
+    r_year = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{year_id}/custom_field_values', json=
+          year_payload
+        , params={
+      }, headers=headers)
+
+    r_client = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{client_name_id}/custom_field_values', json=
+          client_payload
+        , params={
+      }, headers=headers)
+
+    r_project = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{project_name_id}/custom_field_values', json=
+          project_payload
+        , params={
+      }, headers=headers)
+
+    r_caption = requests.post(f'https://brandfolder.com/api/v4/custom_field_keys/{caption_id}/custom_field_values', json=
+        caption_payload
+      , params={
+    }, headers=headers)
+
+    return image_name
+
+
+def run_preprocess_ai(topical_map, client_name, section_type='AI Processed Images', progress=gr.Progress()):
+  section_id = section_key_dict[section_type]
+  headers = {
+    'Content-Type': 'application/json',
+    'Authorization': brandfolder_api
+  }
+  collection_dict = get_collection_dict()
+  collection_id = collection_dict[client_name]
+  page = 1
+  pre_process_ids = []
+  run = True
+  while run == True:
+    r = requests.get(f'https://brandfolder.com/api/v4/collections/{collection_id}/assets?include=section,custom_fields&fields=created_at&page={page}&per=3000&sort_by=created_at&order=DESC', params={}, headers=headers)    
+    page+=1
+    asset_names = [item['id'] for item in r.json()['data'] if item['relationships']['section']['data']['id'] == section_id]
+    if asset_names in pre_process_ids:
+      run = False
+    else:
+        pre_process_ids.append(asset_names)
+  asset_names = sum(pre_process_ids, [])
+  
+  new_imgs = []
+  for asset_id in progress.tqdm(asset_names, desc="Uploading..."):
+    time.sleep(2)
+    asset_dict = get_asset_info(asset_id)
+    new_img = create_ai_asset(asset_dict, topical_map, client_name, new_imgs)
+    new_imgs.append(new_img)
+    
+  gr.Info('Images have been processed!')
+  return