app.py

import gradio as gr
from PIL import Image
from authtoken import auth_token
import torch
import torch.cuda.amp as amp
from diffusers import StableDiffusionPipeline

modelid = "CompVis/stable-diffusion-v1-4"
device = torch.device("cpu")  # Default to CPU device
if torch.cuda.is_available():
    device = torch.device("cuda")
pipe = StableDiffusionPipeline.from_pretrained(modelid, use_auth_token=auth_token)
pipe.to(device)

def generate(prompt):
    with torch.no_grad(), amp.autocast(enabled=device != torch.device("cpu")):
        image = pipe(prompt, guidance_scale=8.5)["sample"][0]

    image.save('generatedimage.png')
    return image

def predict_text(prompt):
    image = generate(prompt)
    return image

def predict_image(input_image):
    input_image.save('input_image.png')
    prompt = input("Enter your prompt: ")
    image = generate(prompt)
    return image

iface = gr.Interface(
    fn=predict_text,
    inputs="text",
    outputs="image",
    capture_session=True,
)
iface.launch()