app.py

import torch
import os
import random
import gradio as gr

from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan, pipeline
import base64
from datasets import load_dataset
from diffusers import DiffusionPipeline
from huggingface_hub import login
import numpy as np
import spaces

@spaces.GPU
def guessanImage(model, image):
    imgclassifier  = pipeline("image-classification", model=model)
    if image is not None:  
        description = imgclassifier(image)
    return description

@spaces.GPU
def guessanAge(model, image):
    imgclassifier  = pipeline("image-classification", model=model)
    if image is not None:  
        description = imgclassifier(image)
    return description    

@spaces.GPU(duration=120)
def text2speech(model, text, voice):
    print(voice)
    if len(text) > 0:
        synthesiser = pipeline("text-to-speech", model=model)
        
        embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
        speaker_embedding = torch.tensor(embeddings_dataset[voice]["xvector"]).unsqueeze(0)
        
        speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
        audio_data = np.frombuffer(speech["audio"], dtype=np.float32)
        audio_data_16bit = (audio_data * 32767).astype(np.int16)
        return speech["sampling_rate"], audio_data_16bit

@spaces.GPU
def ImageGenFromText(text, model):
    api_key = os.getenv("fluxauth")
    login(token=api_key)
    
    if len(text) > 0:
        dtype = torch.bfloat16
        device = "cuda" if torch.cuda.is_available() else "cpu"
        MAX_SEED = np.iinfo(np.int32).max
        seed = random.randint(0, MAX_SEED)
        pipe = DiffusionPipeline.from_pretrained(model, torch_dtype=dtype).to(device)
        generator = torch.Generator().manual_seed(seed)
        image = pipe(
                prompt = text, 
                width = 512,
                height = 512,
                num_inference_steps = 4, 
                generator = generator,
                guidance_scale=0.0
        ).images[0]
        print(image)
        return image


radio1 = gr.Radio(["microsoft/resnet-50", "google/vit-base-patch16-224", "apple/mobilevit-small"], value="microsoft/resnet-50", label="Select a Classifier", info="Image Classifier")
tab1 = gr.Interface(
    fn=guessanImage,
    inputs=[radio1, gr.Image(type="pil")],
    outputs=["text"],
)

radio2 = gr.Radio(["nateraw/vit-age-classifier"], value="nateraw/vit-age-classifier", label="Select an Age Classifier", info="Age Classifier")
tab2 = gr.Interface(
    fn=guessanAge,
    inputs=[radio2, gr.Image(type="pil")],
    outputs=["text"],
)
textbox = gr.Textbox(value="good morning pineapple! looking very good very nice!")
radio3 = gr.Radio(["microsoft/speecht5_tts"], value="microsoft/speecht5_tts", label="Select an tts", info="Age Classifier")
radio3_1 = gr.Radio([("Scottish male (awb)", 0), ("US male (bdl)", 1138), ("US female (clb)", 2271), ("Canadian male (jmk)",3403), ("Indian male (ksp)", 4535), ("US male (rms)", 5667), ("US female (slt)", 6799)], value=4535)
tab3 = gr.Interface(
    fn=text2speech,
    inputs=[radio3, textbox, radio3_1],
    outputs=["audio"],
)

radio4 = gr.Radio(["black-forest-labs/FLUX.1-schnell"], value="black-forest-labs/FLUX.1-schnell", label="Select", info="text to image")
tab4 = gr.Interface(
    fn=ImageGenFromText,
    inputs=["text", radio4],
    outputs=["image"],
)

demo = gr.TabbedInterface([tab1, tab2, tab3, tab4], ["tab1", "tab2", "tab3", "tab4"])
demo.launch()