app.py

import random
import torch
import numpy as np 
import gradio as gr
import spaces

from diffusers import StableDiffusionXLPipeline, AutoencoderKL
from diffusers import DPMSolverMultistepScheduler as DefaultDPMSolver

# Add support for setting custom timesteps
class DPMSolverMultistepScheduler(DefaultDPMSolver):
    def set_timesteps(
        self, num_inference_steps=None, device=None, 
        timesteps=None
    ):
        if timesteps is None:
            super().set_timesteps(num_inference_steps, device)
            return
        
        all_sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
        self.sigmas = torch.from_numpy(all_sigmas[timesteps])
        self.timesteps = torch.tensor(timesteps[:-1]).to(device=device, dtype=torch.int64) # Ignore the last 0
        
        self.num_inference_steps = len(timesteps)

        self.model_outputs = [
            None,
        ] * self.config.solver_order
        self.lower_order_nums = 0

        # add an index counter for schedulers that allow duplicated timesteps
        self._step_index = None
        self._begin_index = None
        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication

vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16, variant="fp16", use_safetensors=True,
    vae=vae,
).to("cuda")

pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

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

@spaces.GPU
def run(prompt="a photo of an astronaut riding a horse on mars",
        negative_prompt="",
        randomize_seed=False,
        seed=20,
        progress=gr.Progress(track_tqdm=True)
    ):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    sampling_schedule = [999, 845, 730, 587, 443, 310, 193, 116, 53, 13, 0]
    torch.manual_seed(seed)
    ays_images = pipe(
        prompt,
        negative_prompt=negative_prompt,
        timesteps=sampling_schedule,
    ).images
    return ays_images[0], seed

examples = [
    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
    "An astronaut riding a green horse",
    "A delicious ceviche cheesecake slice",
]

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

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""
        # Align your steps (AYS) - Stable Diffusion XL
        Unnoficial demo for the official diffusers implementation of the [Align your Steps](https://research.nvidia.com/labs/toronto-ai/AlignYourSteps/) scheduler by NVIDIA for SDXL
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            run_button = gr.Button("Run", scale=0)
        
        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):
            
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
                visible=False,
            )
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
    gr.on(
        [run_button.click, prompt.submit, negative_prompt.submit],
        fn = run,
        inputs = [prompt, negative_prompt, randomize_seed, seed],
        outputs = [result, seed]
    )

demo.launch()