app.py

#!/usr/bin/env python
# coding: utf-8

import os
import openai
import gradio as gr

import torch
from diffusers import StableDiffusionPipeline
from torch import autocast
#from PIL import Image
#from torchvision import transforms

#from diffusers import StableDiffusionImageVariationPipeline


openai.api_key = os.getenv('openaikey')

device = "cuda" if torch.cuda.is_available() else "cpu"
pipe = StableDiffusionPipeline.from_pretrained("stale2000/sd-dnditem", torch_dtype=torch.float16)  
pipe = pipe.to(device)

def predict(input, manual_query_repacement, history=[]):

    # gpt3
    if manual_query_repacement != "":
        input = manual_query_repacement
    
    response = openai.Completion.create(
    model="text-davinci-003",
    prompt=input,
    temperature=0.9,
    max_tokens=150,
    top_p=1,
    frequency_penalty=0,
    presence_penalty=0.6)
    
    # tokenize the new input sentence
    responseText = response["choices"][0]["text"]
    history.append((input, responseText))


    #img generation
    prompt = "Yoda"
    scale = 10
    n_samples = 4

    # Sometimes the nsfw checker is confused by the Naruto images, you can disable
    # it at your own risk here
    #disable_safety = False

    #if disable_safety:
    #  def null_safety(images, **kwargs):
    #      return images, False
    #  pipe.safety_checker = null_safety

    with autocast("cuda"):
      images = pipe(n_samples*[prompt], guidance_scale=scale).images

    for idx, im in enumerate(images):
      im.save(f"{idx:06}.png")

    images_list = pipe(
        inp.tile(n_samples, 1, 1, 1),
        guidance_scale=scale,
        num_inference_steps=steps,
        generator=generator,
        )

    images = []
    for i, image in enumerate(images_list["images"]):
        if(images_list["nsfw_content_detected"][i]):
            safe_image = Image.open(r"unsafe.png")
            images.append(safe_image)
        else:
            images.append(image)


    
    return history, history, images



inputText = gr.Textbox(value="tmp")
manual_query = gr.Textbox(placeholder="Input any query here, to replace the image generation query builder entirely.")

output_img = gr.Gallery(label="Generated image")
output_img.style(grid=2)

gr.Interface(fn=predict,
             inputs=[inputText,manual_query,'state'],
            
             outputs=["chatbot",'state', output_img]).launch()