app.py

import torch
import gradio as gr

from huggingface_hub import hf_hub_download
from safetensors.torch import load_file

from models.unet import PixelArtUNet
from sampling.conditional_probability_path import GaussianConditionalProbabilityPath
from sampling.noise_scheduling import LinearAlpha, LinearBeta
from diff_eq.ode_sde import UnguidedVectorFieldODE
from diff_eq.simulator import EulerSimulator
from utils import tensor_to_rgba_image, normalize_to_unit, make_large

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

# Setup model
model = PixelArtUNet(
    channels = [128, 256, 512, 1024],
    num_residual_layers = 2,
    t_embed_dim = 128,
    midcoder_dropout_p=0.2
).to(device)

repo_id = "mradovic38/sprite-flow"
filename = "model.safetensors"
file_path = hf_hub_download(repo_id=repo_id, filename=filename)
checkpoint = load_file(file_path)
model.load_state_dict(checkpoint)
model.to(device)
model.eval()


def generate_image_stream(num_timesteps: int = 200):
    # Setup path
    path = GaussianConditionalProbabilityPath(
        p_data=None,
        p_simple_shape=[4, 128, 128],
        alpha=LinearAlpha(),
        beta=LinearBeta()
    ).to(device)
    path.eval()
    ts = torch.linspace(0, 1, num_timesteps).view(1, -1, 1, 1, 1).expand(1, -1, 1, 1, 1).to(device)
    x0 = path.p_simple.sample(1).to(device)  # (1, 4, 128, 128)
    ode = UnguidedVectorFieldODE(model)
    simulator = EulerSimulator(ode)

    # Yield images at each step
    for x in simulator.simulate(x0, ts):
        img = normalize_to_unit(x)
        img = tensor_to_rgba_image(img)[0]
        yield make_large(img)


# --- Create Gradio interface ---
css = """
.gradio-container {
    max-width: 700px !important;
    margin: 0 auto !important;  /* centers the container */
    text-align: center;          /* centers text and components inside */
}

#component-0 {
    display: inline-block;       /* make the image inline-block */
    width: 100% !important;
}
"""
iface = gr.Interface(
    fn=generate_image_stream,
    inputs=[gr.Slider(50, 500, step=10, value=200, label="Number of Steps")],
    outputs=gr.Image(type="pil", streaming=True),
    title="Flow-Matching Pixel Art Sprite Generation",
    description="Generate pixel art sprites with sprite-flow.",
    css=css
)

if __name__ == "__main__":
    iface.launch()