app.py

import gradio as gr
import numpy as np
import random
#from diffusers import DiffusionPipeline
from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
#from diffusers.utils import load_image
import torch
from PIL import Image
from huggingface_hub import hf_hub_download
from eSeNTranslate import TranslateFromAny2XModel

# Load translation model(s) (Pipeline)
fasttextModelPath = hf_hub_download(repo_id="facebook/fasttext-language-identification", filename="model.bin")
translatePipe = TranslateFromAny2XModel(nllb_model_path="facebook/nllb-200-distilled-600M", fasttext_model_path=fasttextModelPath)

modelPath = "stabilityai/sdxl-turbo"
 
if torch.cuda.is_available():
    device = "cuda"
    torch.cuda.max_memory_allocated(device=device)
    #pipe = DiffusionPipeline.from_pretrained(modelPath, torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
    pipeTex2Image = AutoPipelineForText2Image.from_pretrained(modelPath, torch_dtype=torch.float16, variant="fp16")
    pipeImage2Image = AutoPipelineForImage2Image.from_pretrained(modelPath, torch_dtype=torch.float16, variant="fp16")
    #pipe.enable_xformers_memory_efficient_attention()
    pipeTex2Image.enable_xformers_memory_efficient_attention() 
    pipeImage2Image.enable_xformers_memory_efficient_attention()
    #pipe = pipe.to(device)
else:
    device = "cpu"
    #pipe = DiffusionPipeline.from_pretrained(modelPath, use_safetensors=True)
    pipeTex2Image = AutoPipelineForText2Image.from_pretrained(modelPath, use_safetensors=True)
    pipeImage2Image = AutoPipelineForImage2Image.from_pretrained(modelPath, use_safetensors=True)
    #pipe = pipe.to(device)

#pipe = pipe.to(device)    
pipeTex2Image.to(device)
pipeImage2Image.to(device)

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

def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, use_as_input, strength, image):

    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
        
    generator = torch.Generator().manual_seed(seed)
    prompt = translatePipe.generate(prompt)
    
    if use_as_input:
        print("Image to Image:")
        pipe = pipeImage2Image
        init_image = Image.fromarray(np.uint8(image)).resize((width, height)).convert("RGB")
        init_image.save("input.png", format="PNG")
        print(type(init_image), init_image.size)
        image = pipe(
            prompt = prompt, 
            negative_prompt = negative_prompt,
            guidance_scale = guidance_scale, 
            num_inference_steps = num_inference_steps, 
            width = width, 
            height = height,
            generator = generator,
            strength=strength,
            image=init_image
        ).images[0] 
    else:
        print("Text to Image:")
        pipe = pipeTex2Image
        image = pipe(
            prompt = prompt, 
            negative_prompt = negative_prompt,
            guidance_scale = guidance_scale, 
            num_inference_steps = num_inference_steps, 
            width = width, 
            height = height,
            generator = generator
        ).images[0] 
    
    return image

examples = [
    "Face  of a modern woman of Balkan descent 25 years old",
    "Blue car sandero stepway on dirt road",
    "Cow in the skin of a dog of dalmatian breed",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: auto;
}
"""


with gr.Blocks(css=css) as app:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""
        # Text-to-Image, Image-to-Image by Slavko Novak
        Currently running on {device}.
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            run_button = gr.Button("Generate", scale=0)
        
        result = gr.Image(label="Result", show_label=False)
        use_as_input = gr.Checkbox(label="Use image as input", value=False)

        with gr.Accordion("Advanced Settings", open=False):
            
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
                visible=True,
            )
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=10.0,
                    step=0.1,
                    value=0.0,
                )
                
                strength = gr.Slider(
                    label="Strength scale",
                    minimum=0.0,
                    maximum=1.0,
                    step=0.1,
                    value=0.5,
                )
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=12,
                    step=1,
                    value=2,
                )
        
        gr.Examples(
            examples = examples,
            inputs = [prompt]
        )

    run_button.click(
        fn = infer,
        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, use_as_input, strength, result],
        outputs = [result]
    )

#app.queue().launch(server_name="0.0.0.0", server_port=8080, share=True)
app.queue().launch()