app.py

import gradio as gr
import torch
from diffusers import StableDiffusionXLPipeline, AutoencoderKL
from huggingface_hub import hf_hub_download
import spaces
from PIL import Image
import requests
from translatepy import Translator

translator = Translator()

# Constants
model = "Corcelio/openvision"
vae_model = "madebyollin/sdxl-vae-fp16-fix"

CSS = """
.gradio-container {
  max-width: 690px !important;
}
footer {
    visibility: hidden;
}
"""

JS = """function () {
  gradioURL = window.location.href
  if (!gradioURL.endsWith('?__theme=dark')) {
    window.location.replace(gradioURL + '?__theme=dark');
  }
}"""

# Load VAE component
vae = AutoencoderKL.from_pretrained(
    vae_model, 
    torch_dtype=torch.float16
)

# Ensure model and scheduler are initialized in GPU-enabled function
if torch.cuda.is_available():
    pipe = StableDiffusionXLPipeline.from_pretrained(model, vae=vae, torch_dtype=torch.float16).to("cuda")



# Function 
@spaces.GPU()
def generate_image(
    prompt,
    negative,
    width=1024,
    height=1024,
    scale=1.5,
    steps=30):
    
    prompt = str(translator.translate(prompt, 'English'))

    print(f'prompt:{prompt}')
    
    image = pipe(
        prompt, 
        negative_prompt=negative, 
        width=width,
        height=height,
        guidance_scale=scale,
        num_inference_steps=steps,
    )
    print(image.images[0])
    return image.images[0]


examples = [
    "a cat eating a piece of cheese",
    "a ROBOT riding a BLUE horse on Mars, photorealistic",
    "Ironman VS Hulk, ultrarealistic",
    "a CUTE robot artist painting on an easel",
    "Astronaut in a jungle, cold color palette, oil pastel, detailed, 8k",
    "An alien holding sign board contain word 'Flash', futuristic, neonpunk",
    "Kids going to school, Anime style"
]


# Gradio Interface

with gr.Blocks(css=CSS, js=JS, theme="soft") as demo:
    gr.HTML("<h1><center>OpenVision</center></h1>")
    gr.HTML("<p><center><a href='https://huggingface.co/Corcelio/openvision'>OpenVision</a> text-to-image generation</center><br><center>Multi-Languages. Midjourney Aesthetic for All Your Images</center></p>")
    with gr.Group():
        with gr.Row():
            prompt = gr.Textbox(label='Enter Your Prompt', scale=6)
            submit = gr.Button(scale=1, variant='primary')
    img = gr.Image(label='OpenVision Generated Image')
    with gr.Accordion("Advanced Options", open=False):
        with gr.Row():
            negative = gr.Textbox(label="Negative prompt", value="low quality")
        with gr.Row():
            width = gr.Slider(
                label="Width",
                minimum=512,
                maximum=1280,
                step=8,
                value=1024,
            )
            height = gr.Slider(
                label="Height",
                minimum=512,
                maximum=1280,
                step=8,
                value=1024,
            )
        with gr.Row():
            scale = gr.Slider(
                label="Guidance Scale",
                minimum=0,
                maximum=50,
                step=0.1,
                value=1.5,
            )
            steps = gr.Slider(
                label="Steps",
                minimum=1,
                maximum=50,
                step=1,
                value=30,
            )            
    gr.Examples(
        examples=examples,
        inputs=[prompt, negative, width, height, scale, steps],
        outputs=img,
        fn=generate_image,
        cache_examples="lazy",
    )

    prompt.submit(fn=generate_image,
                 inputs=[prompt, negative, width, height, scale, steps],
                 outputs=img,
                 )
    submit.click(fn=generate_image,
                 inputs=[prompt, negative, width, height, scale, steps],
                 outputs=img,
                 )
    
demo.queue().launch()