app.py

#!/usr/bin/env python

import pathlib
import pickle
import sys

import gradio as gr
import numpy as np
import torch
from huggingface_hub import hf_hub_download
from torch import nn

sys.path.insert(0, "stylegan3")

TITLE = "StyleGAN3 Anime Face Generation"

MODEL_REPO = "hysts/stylegan3-anime-face-exp002-model"
MODEL_FILE_NAME = "009000.pkl"


def make_transform(translate: tuple[float, float], angle: float) -> np.ndarray:
    mat = np.eye(3)
    sin = np.sin(angle / 360 * np.pi * 2)
    cos = np.cos(angle / 360 * np.pi * 2)
    mat[0][0] = cos
    mat[0][1] = sin
    mat[0][2] = translate[0]
    mat[1][0] = -sin
    mat[1][1] = cos
    mat[1][2] = translate[1]
    return mat


def load_model(device: torch.device) -> nn.Module:
    path = hf_hub_download(MODEL_REPO, MODEL_FILE_NAME)
    with pathlib.Path(path).open("rb") as f:
        model = pickle.load(f)  # noqa: S301
    model.eval()
    model.to(device)
    with torch.inference_mode():
        z = torch.zeros((1, 512)).to(device)
        c = torch.zeros(0).to(device)
        model(z, c)
    return model


device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = load_model(device)


def generate_z(seed: int, device: torch.device) -> torch.Tensor:
    return torch.from_numpy(np.random.RandomState(seed).randn(1, 512)).to(device)


@torch.inference_mode()
def generate_image(seed: int, truncation_psi: float, tx: float, ty: float, angle: float) -> np.ndarray:
    seed = int(np.clip(seed, 0, np.iinfo(np.uint32).max))
    z = generate_z(seed, device)
    c = torch.zeros(0).to(device)

    mat = make_transform((tx, ty), angle)
    mat = np.linalg.inv(mat)
    model.synthesis.input.transform.copy_(torch.from_numpy(mat))

    out = model(z, c, truncation_psi=truncation_psi)
    out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(torch.uint8)
    return out[0].cpu().numpy()


demo = gr.Interface(
    fn=generate_image,
    inputs=[
        gr.Slider(label="Seed", minimum=0, maximum=np.iinfo(np.uint32).max, step=1, value=3407851645),
        gr.Slider(label="Truncation psi", minimum=0, maximum=2, step=0.05, value=0.7),
        gr.Slider(label="Translate X", minimum=-1, maximum=1, step=0.05, value=0),
        gr.Slider(label="Translate Y", minimum=-1, maximum=1, step=0.05, value=0),
        gr.Slider(label="Angle", minimum=-180, maximum=180, step=5, value=0),
    ],
    outputs=gr.Image(label="Output"),
    title=TITLE,
    css_paths="style.css",
)

if __name__ == "__main__":
    demo.queue(max_size=20).launch()