.env usage

.gitignore

@@ -1,4 +1,5 @@
 __pycache__/
 *.pyc
-OPENAI_API_KEY.txt
-replicate_api.txt
+.env
+Data/PngRealSet
+Data/version2.csv

modules/im2tx.py

@@ -3,6 +3,7 @@
 import time
 import pandas as pd
 from PIL import Image
+import torch
 from transformers import BlipProcessor as Bp
 from transformers import BlipForConditionalGeneration as BConGen
 from tqdm import tqdm
@@ -21,10 +22,11 @@ def obtain_df_with_text(data_path : str, verbose : bool=False ) -> pd.DataFrame:
     """
     Given a path to a folder with images, returns a DataFrame with the images and their descriptions.
     """
+    device = "cuda" if torch.cuda.is_available() else "cpu"
     warnings.simplefilter(action='ignore', category=FutureWarning)
     processor = Bp.from_pretrained("Salesforce/blip-image-captioning-large")
     # Add .to("cuda") to run in GPU
-    model = BConGen.from_pretrained("Salesforce/blip-image-captioning-large")
+    model = BConGen.from_pretrained("Salesforce/blip-image-captioning-large").to(device)
     # We use this text to condition the image text generation
     seed1 = "a photography of"
     # We store here the images and their descriptions
@@ -36,7 +38,7 @@ def obtain_df_with_text(data_path : str, verbose : bool=False ) -> pd.DataFrame:
             img_path = os.path.join(data_path, img).replace("\\", "/")
             raw_image = Image.open(img_path).convert('RGB')
             # Add .to("cuda") to run in GPU
-            inputs = processor(raw_image, seed1, return_tensors="pt")
+            inputs = processor(raw_image, seed1, return_tensors="pt").to(device)
             out = model.generate(**inputs, max_new_tokens=20)
             capt = processor.decode(out[0], skip_special_tokens=True)
             imtx.loc[len(imtx.index)] = [img_path, capt]

modules/iris_db.py

@@ -118,21 +118,15 @@ def description_search(self, query: str, top_n: int, name: str = 'gallery_images
 
 
 if __name__ == '__main__':
-    OPENAI_API_KEY = open('OPENAI_API_KEY.txt').read().strip()
-    iris = IrisDB(Openai_client = OpenAI(api_key=OPENAI_API_KEY))
+    iris = IrisDB(Openai_client = OpenAI())
 
-    if not iris.table_exists():
-        # df = pd.DataFrame({
-        #     'image_path': ['image1.jpg', 'image2.jpg', 'image3.jpg'],
-        #     'description': ['a cat', 'a dog', 'a car']
-        # })
-        df = pd.read_csv('Data/version2.csv')
-        df.columns = ['image_path', 'description']
+    df = pd.read_csv('Data/version2.csv')
+    df.columns = ['image_path', 'description']
 
-        iris.init_table()
-        iris.insert_df_to_table(df)
+    iris.init_table()
+    iris.insert_df_to_table(df)
 
-    description_search = "music"
+    description_search = "beard"
     results = iris.description_search(description_search, top_n=3)
     # limit prints of pd.DataFrame to 240 characters
     pd.set_option('display.max_colwidth', 240)

modules/manager.py

@@ -5,6 +5,9 @@
 from openai import OpenAI
 import pandas as pd
 import os
+from dotenv import load_dotenv
+
+load_dotenv()
 
 def generate_videoclip(
         openai_client : OpenAI,
@@ -18,7 +21,7 @@ def generate_videoclip(
     Given a query, generates a video clip with images and a song.
     """
     if not os.environ.get("REPLICATE_API_TOKEN"):
-        os.environ["REPLICATE_API_TOKEN"] = open('replicate_api.txt').read().strip()
+        raise Exception("Please set the REPLICATE_API_TOKEN environment variable.")
 
     db_output_df  = DB.description_search(query=query, top_n=amount_images)
     image_paths = db_output_df['path'].tolist()
@@ -31,8 +34,7 @@ def generate_videoclip(
 
 
 if __name__ == '__main__':
-    OPENAI_API_KEY = open('OPENAI_API_KEY.txt').read().strip()
-    Openai_client = OpenAI(api_key=OPENAI_API_KEY)
+    Openai_client = OpenAI()
     DB = IrisDB(Openai_client = Openai_client)
     if not DB.table_exists():
         df = pd.read_csv('Data/version2.csv')
@@ -42,4 +44,4 @@ def generate_videoclip(
         DB.insert_df_to_table(df)
 
     query = 'Cool chairs'
-    generate_videoclip(Openai_client, DB, query, duration=10, video_path='Data/ExampleData/output4.mp4', amount_images=6)
+    generate_videoclip(Openai_client, DB, query, duration=10, video_path='Data/ExampleData/output.mp4', amount_images=6)

modules/song_prompt_creator.py

@@ -36,7 +36,7 @@ def create_song_prompt(query : str, image_descriptions : list[str], openai_clien
 if __name__ == '__main__':
     if not os.environ.get("OPENAI_API_KEY"): 
         os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")
-    openai_client = OpenAI(api_key=os.getenv('OPENAI_API_KEY'))
+    openai_client = OpenAI()
     query = "Sadness"
     image_descriptions = ["Long winter nights", "A lonely person", "A rainy day"]
     print(create_song_prompt(query, image_descriptions, openai_client))