app.py

import gradio as gr
import spaces
import numpy as np
import random
from diffusers import DiffusionPipeline
import torch
import threading
from PIL import Image

MODEL_ID = "cagliostrolab/animagine-xl-3.1"
device = "cuda" if torch.cuda.is_available() else "cpu"

if torch.cuda.is_available():
    torch.cuda.max_memory_allocated(device=device)
    pipe = DiffusionPipeline.from_pretrained(
        MODEL_ID, 
        torch_dtype=torch.float16, 
        use_safetensors=True, 
    )
else:
    pipe = DiffusionPipeline.from_pretrained(MODEL_ID, use_safetensors=True)
pipe = pipe.to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1536

def latents_to_rgb(latents):
    weights = (
        (60, -60, 25, -70),
        (60, -5, 15, -50),
        (60, 10, -5, -35)
    )

    weights_tensor = torch.tensor(weights, dtype=latents.dtype, device=latents.device).T
    biases_tensor = torch.tensor((150, 140, 130), dtype=latents.dtype, device=latents.device)
    rgb_tensor = torch.einsum("...lxy,lr -> ...rxy", latents, weights_tensor) + biases_tensor.view(-1, 1, 1)
    image_array = rgb_tensor.clamp(0, 255)[0].byte().cpu().numpy()
    image_array = image_array.transpose(1, 2, 0)  # Change the order of dimensions

    pil_image = Image.fromarray(image_array)

    resized_image = pil_image.resize((pil_image.size[0] * 2, pil_image.size[1] * 2), Image.LANCZOS)  # Resize 128x128 * ...
    return resized_image

class BaseGenerator:
    def __init__(self, pipe):
        self.pipe = pipe
        self.image = None
        self.new_image_event = threading.Event()
        self.generation_finished = threading.Event()
        self.intermediate_image_concurrency(3)

    def intermediate_image_concurrency(self, concurrency):
        self.concurrency = concurrency
    
    def decode_tensors(self, pipe, step, timestep, callback_kwargs):
        latents = callback_kwargs["latents"]
        if step % self.concurrency == 0:  # every how many steps
            print(step)
            self.image = latents_to_rgb(latents)
            self.new_image_event.set()  # Signal that a new image is available
        return callback_kwargs

    def show_images(self):
        while not self.generation_finished.is_set() or self.new_image_event.is_set():
            self.new_image_event.wait()  # Wait for a new image
            self.new_image_event.clear()  # Clear the event flag

            if self.image:
                yield self.image  # Yield the new image

    def generate_images(self, **kwargs):
        if kwargs.get('randomize_seed', False):
            kwargs['seed'] = random.randint(0, MAX_SEED)
            
        generator = torch.Generator().manual_seed(kwargs['seed'])
    
        self.image = None
        self.image = self.pipe(
            height=kwargs['height'],
            width=kwargs['width'],
            prompt=kwargs['prompt'],
            negative_prompt=kwargs['negative_prompt'],
            guidance_scale=kwargs['guidance_scale'],
            num_inference_steps=kwargs['num_inference_steps'],
            generator=generator,
            callback_on_step_end=self.decode_tensors,
            callback_on_step_end_tensor_inputs=["latents"],
        ).images[0]
        print("finish")
        self.new_image_event.set()  # Result image
        self.generation_finished.set()  # Signal that generation is finished

    def stream(self, prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
        self.generation_finished.clear()
        threading.Thread(target=self.generate_images, args=(), kwargs=dict(
            prompt=prompt,
            negative_prompt=negative_prompt,
            seed=seed,
            randomize_seed=randomize_seed,
            width=width,
            height=height,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps
        )).start()
        return self.show_images()

image_generator = BaseGenerator(pipe)

@spaces.GPU
def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, concurrency):

    image_generator.intermediate_image_concurrency(concurrency)
    
    stream = image_generator.stream(
        prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps
    )

    yield None
    
    for image in stream:
        yield image


css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

if torch.cuda.is_available():
    power_device = "GPU"
else:
    power_device = "CPU"

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""
        # Text-to-Image: Display each generation step
        
        Gradio template for displaying preview images during generation steps
        
        Currently running on {power_device}.
        """)

        prompt = gr.Text(
            label="Prompt",
            show_label=False,
            max_lines=1,
            placeholder="Enter your prompt",
            container=False,
            value="1girl, souryuu asuka langley, neon genesis evangelion, solo, upper body, v, smile, looking at viewer, outdoors, night",
        )

        negative_prompt = gr.Text(
            label="Negative prompt",
            max_lines=1,
            placeholder="Enter a negative prompt",
            visible=True,
            value="nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]",
        )
        
        with gr.Row():
            
            run_button = gr.Button("Run", scale=0)
        
        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):
            
            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=832,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1216,
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=30.0,
                    step=0.1,
                    value=7.0,
                )
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=100,
                    step=1,
                    value=76,
                )

            concurrency_gui = gr.Slider(
                label="Number of steps to show the next preview image",
                minimum=1,
                maximum=20,
                step=1,
                value=3,
            )

    run_button.click(
        fn = infer,
        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, concurrency_gui],
        outputs = [result],
        show_progress="minimal",
    )

demo.queue().launch()