app.py

import torch
import PIL.Image
import gradio as gr
import numpy as np

from .pipeline import DDIMPipelineCustom

pipeline = DDIMPipelineCustom.from_pretrained("1aurent/ddpm-mnist-conditional")

def predict(steps, seed, value, guidance):
    generator = torch.manual_seed(seed)
    for i in range(1,steps):
        yield pipeline(
            generator=generator,
            condition=torch.tensor([value]),
            guidance=guidance,
            num_inference_steps=steps
        ).images[0]

gr.Interface(
    predict,
    inputs=[
        gr.components.Slider(1, 100, label='Inference Steps', value=20, step=1),
        gr.components.Slider(0, 2147483647, label='Seed', value=69420, step=1),
        gr.components.Slider(0, 9, label='Value', value=5, step=1),
        gr.components.Slider(-2.5, 2.5, label='Guidance Factor', value=1),
    ],
    outputs=gr.Image(shape=[28,28], type="pil", elem_id="output_image"),
    css="#output_image img {width: 256px}",
    title="Conditional MNIST",
    description="A DDIM scheduler and UNet model trained on the MNIST dataset for conditional image generation.",
).queue().launch()