app.py

from transformers import pipeline, Conversation
import gradio as gr
from diffusers import DiffusionPipeline
import scipy

#Initializing Models
chatbot = pipeline(model="facebook/blenderbot-400M-distill")
ldm = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
synthesiser = pipeline("text-to-audio", "facebook/musicgen-small")

message_list = []
response_list = []
def vanilla_chatbot(message):
    conversation = Conversation(text=message, past_user_inputs=message_list, generated_responses=response_list)
    bot = chatbot(conversation.messages[0]['content']) # working code
    return bot[-1]['generated_text']

def generate_image(Prompt):
    images = ldm([Prompt], num_inference_steps=50, eta=.3, guidance_scale=6)
    return images.images[0]

def generate_music(Prompt):
    music = synthesiser(Prompt, forward_params={"do_sample": True, "max_new_tokens":100})
    rate = music["sampling_rate"]
    mus = music["audio"][0].reshape(-1)
    return rate,mus

def process_input(Prompt,choice):
    if choice == "Chat":
        return vanilla_chatbot(Prompt),None,None
    elif choice == 'Music':
        rate,audio = generate_music(Prompt)
        return None, (rate,audio), None
    else:
        return None , None , generate_image(Prompt)

# demo=gr.Blocks()
# with demo:
#     with gr.Row():
#         text_input = gr.Textbox()
#         choice = gr.Radio(choices=["Chat","Music","Image"])

#     with gr.Row():
#         chatbot_output = gr.Textbox()
#         music_output =gr.Audio()
#         image_output =gr.Image()
        
#     submit_btn = gr.Button("Generate")
    
#     submit_btn.click(fn=process_input,inputs=[text_input,choice],outputs=[chatbot_output,music_output,image_output])

# demo.launch(debug=True)


demo =gr.Interface(
    fn=process_input,
    inputs=[gr.Textbox(),gr.Radio(["Chat","Music","Image"])],
    outputs = [gr.Textbox(),gr.Audio(),gr.Image()],
    # outputs =["text","audio","image"]
    title="Multimodal Assistant"

)

demo.launch(debug=True)