app.py

import gradio as gr
import numpy as np
import random
from diffusers import DiffusionPipeline
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

model_turbo_repo_id = "stabilityai/sdxl-turbo"
model_repo_id = "stabilityai/stable-diffusion-2"
pipe = DiffusionPipeline.from_pretrained(model_turbo_repo_id, torch_dtype=torch_dtype)
pipe = pipe.to(device)

# Recommended if your computer has < 64 GB of RAM
pipe.enable_attention_slicing()

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024


def create_generator(seed):
    return torch.Generator().manual_seed(seed)


def generate(prompt, negative_prompt, seed, is_random_seed, width, height, guidance_scale, inference_steps):
    if is_random_seed:
        seed = random.randint(0, np.iinfo(np.int32).max)

    generator = create_generator(seed)

    return pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=inference_steps,
        width=width,
        height=height,
        generator=generator
    ).images[0]


with gr.Blocks() as imageGenerator:
    with gr.Column():
        gr.Markdown(f"""
        # Zatsit Image Generator
        ## Générateur d'image basé sur des modèles de stable diffusion
        Vous cherchez de l'inspiration pour vos prompts ? 
        [lien](https://stablediffusion.fr/prompts)
        """)

        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=4,
                placeholder="Votre prompt",
                container=False,
            )

            generate_btn = gr.Button("Générer", scale=0)

        result = gr.Image(label="Image générée", show_label=False)

        with gr.Accordion("Paramètres", open=True):
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=4,
                placeholder="Votre prompt négatif",
            )

            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=70537634,
            )

            is_random_seed = gr.Checkbox(label="Seed en mode random", value=False)

            width = gr.Slider(
                label="Width",
                minimum=256,
                maximum=MAX_IMAGE_SIZE,
                step=10,
                value=1024,
            )

            height = gr.Slider(
                label="Height",
                minimum=256,
                maximum=MAX_IMAGE_SIZE,
                step=10,
                value=1024,
            )

            guidance_scale = gr.Slider(
                label="Guildance scale",
                minimum=0,
                maximum=10.0,
                step=0.1,
                value=0.0,
            )

            inference_steps = gr.Slider(
                label="Nombre d'inférences",
                minimum=0,
                maximum=100.0,
                step=1,
                value=2,
            )

        gr.Examples(
            examples=[
                ["pikachu eating spagetti, Antonio J. Manzanedo", 3384976558, 7, 20],
                ["Gandalf from Lord of the Rings, diffuse lighting, fantasy, intricate elegant highly detailed "
                 "lifelike photorealistic digital painting, artstation", 248215544, 7, 42],
                ["Ethereal gardens of marble built in a shining teal river in future city, gorgeous ornate "
                 "multi-tiered fountain, futuristic, intricate elegant highly detailed lifelike photorealistic "
                 "realistic  painting, long shot, studio lighting, octane render, by Dorian Cleavenger", 3868142022,
                 7, 20],
                ["Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", 70537634, 1, 2]
            ],
            inputs=[prompt, seed, guidance_scale, inference_steps]
        )

    gr.on(
        triggers=[generate_btn.click, prompt.submit],
        fn=generate,
        inputs=[prompt, negative_prompt, seed, is_random_seed, width, height, guidance_scale, inference_steps],
        outputs=[result]
    )

imageGenerator.queue().launch()