app.py

# # %%bash

# # # git lfs install
# # # git clone https://huggingface.co/spaces/Xhaheen/meme_world


# # # pip install -r /content/meme_world/requirements.txt
# # # pip install gradio
# # cd /meme_world


# import torch 
# import re 
# import gradio as gr
# from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel 
# import cohere
# import os
# # 
# # os.environ['key_srkian'] = ''
# key_srkian = os.environ["key_srkian"]
# co = cohere.Client(key_srkian)#srkian
# device='cpu'
# encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
# decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
# model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
# feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
# tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
# model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)


# def predict(department,image,max_length=64, num_beams=4):
#   image = image.convert('RGB')
#   image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
#   clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]
#   caption_ids = model.generate(image, max_length = max_length)[0]
#   caption_text = clean_text(tokenizer.decode(caption_ids))
#   dept=department
#   context= caption_text
#   response = co.generate(
#   model='large',
#   prompt=f'create non offensive one line meme for given department and context\n\ndepartment- data science\ncontext-a man sitting on a bench with a laptop\nmeme- \"I\'m not a data scientist, but I play one on my laptop.\"\n\ndepartment-startup\ncontext-a young boy is smiling while using a laptop\nmeme-\"When your startup gets funded and you can finally afford a new laptop\"\n\ndepartment- {dept}\ncontext-{context}\nmeme-',
#   max_tokens=20,
#   temperature=0.8,
#   k=0,
#   p=0.75,
#   frequency_penalty=0,
#   presence_penalty=0,
#   stop_sequences=["department"],
#   return_likelihoods='NONE')
#   reponse=response.generations[0].text
#   reponse = reponse.replace("department", "")
#   Feedback_SQL="DEPT"+dept+"CAPT"+caption_text+"MAMAY"+reponse
  
     
#   return reponse 



# # input = gr.inputs.Image(label="Upload your Image", type = 'pil', optional=True)
 
 
 
# output = gr.outputs.Textbox(type="text",label="Meme")
# #examples = [f"example{i}.jpg" for i in range(1,7)]
# #examples = os.listdir() 
# examples = [f"example{i}.png" for i in range(1,7)]

# #examples=os.listdir()
# #for fichier in examples:
#   #  if not(fichier.endswith(".png")):
#    #     examples.remove(fichier)
        
# description= " Looking for a fun and easy way to generate memes? Look no further than Meme world! Leveraging large language models like GPT-3PT-3  / Ai21 / Cohere, you can create memes that are sure to be a hit with your friends or network. Created with ♥️ by Arsalan @[Xaheen](https://www.linkedin.com/in/sallu-mandya/). kindly share your thoughts in discussion session and use the app responsibly #NO_Offense \n \n built with ❤️   @[Xhaheen](https://www.linkedin.com/in/sallu-mandya/)"
# title = "Meme world 🖼️"
# dropdown=["data science", "product management","marketing","startup" ,"agile","crypto" , "SEO" ]

# article = "Created By :  Xaheen "

# interface = gr.Interface(
#         fn=predict,
#         inputs = [gr.inputs.Dropdown(dropdown),gr.inputs.Image(label="Upload your Image", type = 'pil', optional=True)],
        
#         theme="grass",
#         outputs=output,
#         examples =[['data science', 'example5.png'],
#             ['product management', 'example2.png'],
#             ['startup', 'example3.png'],
#             ['marketing', 'example4.png'],
#             ['agile', 'example1.png'],
#             ['crypto', 'example6.png']],
#         title=title,
#         description=description,
#         article = article,
#     )
# interface.launch(debug=True)











# Step 2: Set up the Gradio interface and import necessary packages
import gradio as gr
import openai
from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
import torch
from PIL import Image

# Step 3: Load the provided image captioning model
model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
feature_extractor = ViTImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
tokenizer = AutoTokenizer.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Step 4: Create a function to generate captions from images
max_length = 16
num_beams = 4
gen_kwargs = {"max_length": max_length, "num_beams": num_beams}

def generate_caption(image):
    image = Image.fromarray(image.astype('uint8'), 'RGB')
    if image.mode != "RGB":
        image = image.convert(mode="RGB")
    pixel_values = feature_extractor(images=[image], return_tensors="pt").pixel_values
    pixel_values = pixel_values.to(device)
    output_ids = model.generate(pixel_values, **gen_kwargs)
    caption = tokenizer.decode(output_ids[0], skip_special_tokens=True).strip()
    return caption


# Step 5: Create a function to generate memes using the GPT-3 API
def generate_meme(caption, department):
    openai.api_key = os.environ["key"]
    prompt = f"Create a non-offensive meme caption for the following image description in the context of {department} department: {caption}"
    response = openai.Completion.create(engine="text-davinci-002", prompt=prompt, max_tokens=50, n=1, stop=None, temperature=0.7)
    meme_caption = response.choices[0].text.strip()
    return meme_caption

# Step 6: Define the main meme generation function
def meme_generator(image, department):
    caption = generate_caption(image)
    meme_caption = generate_meme(caption, department)
    return meme_caption

examples = [f"example{i}.png" for i in range(1,7)]

# Step 7: Launch the Gradio application
image_input = gr.inputs.Image()
department_input = gr.inputs.Dropdown(choices=["data science", "product management","marketing","startup" ,"agile","crypto" , "SEO" ])
output_text = gr.outputs.Textbox()

gr.Interface(fn=meme_generator, inputs=[image_input, department_input], outputs=output_text, title="Meme world!",description= " Looking for a fun and easy way to generate memes? Look no further than Meme world! Leveraging large language models like GPT-3PT-3  / Ai21 / Cohere, you can create memes that are sure to be a hit with your friends or network. Created with ♥️ by Arsalan @[Xaheen](https://www.linkedin.com/in/sallu-mandya/). kindly share your thoughts in discussion session and use the app responsibly #NO_Offense \n \n built with ❤️   @[Xhaheen](https://www.linkedin.com/in/sallu-mandya/)",        theme="grass",
    
        examples =[['example5.png','data science' ],
            ['example2.png','product management'],
            ['example3.png','startup'],
            ['example4.png','marketing'],
            ['example1.png','agile'],
            ['example6.png','crypto']]).launch(debug=True)