app.py

#!/usr/bin/env python

from __future__ import annotations

import os
import random

import gradio as gr
import numpy as np
import PIL.Image
import spaces
import torch
from diffusers import StableDiffusionAttendAndExcitePipeline, StableDiffusionPipeline

DESCRIPTION = """\
# Attend-and-Excite

This is a demo for [Attend-and-Excite](https://arxiv.org/abs/2301.13826).
Attend-and-Excite performs attention-based generative semantic guidance to mitigate subject neglect in Stable Diffusion.
Select a prompt and a set of indices matching the subjects you wish to strengthen (the `Check token indices` cell can help map between a word and its index).
"""

if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"

if torch.cuda.is_available():
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model_id = "CompVis/stable-diffusion-v1-4"
    ax_pipe = StableDiffusionAttendAndExcitePipeline.from_pretrained(model_id)
    ax_pipe.to(device)
    sd_pipe = StableDiffusionPipeline.from_pretrained(model_id)
    sd_pipe.to(device)


MAX_INFERENCE_STEPS = 100
MAX_SEED = np.iinfo(np.int32).max


def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed


def get_token_table(prompt: str) -> list[tuple[int, str]]:
    tokens = [ax_pipe.tokenizer.decode(t) for t in ax_pipe.tokenizer(prompt)["input_ids"]]
    tokens = tokens[1:-1]
    return list(enumerate(tokens, start=1))


@spaces.GPU
def run(
    prompt: str,
    indices_to_alter_str: str,
    seed: int = 0,
    apply_attend_and_excite: bool = True,
    num_inference_steps: int = 50,
    guidance_scale: float = 7.5,
    scale_factor: int = 20,
    thresholds: dict[int, float] = {
        10: 0.5,
        20: 0.8,
    },
    max_iter_to_alter: int = 25,
) -> PIL.Image.Image:
    if num_inference_steps > MAX_INFERENCE_STEPS:
        raise gr.Error(f"Number of steps cannot exceed {MAX_INFERENCE_STEPS}.")

    generator = torch.Generator(device=device).manual_seed(seed)
    if apply_attend_and_excite:
        try:
            token_indices = list(map(int, indices_to_alter_str.split(",")))
        except Exception:
            raise ValueError("Invalid token indices.")
        out = ax_pipe(
            prompt=prompt,
            token_indices=token_indices,
            guidance_scale=guidance_scale,
            generator=generator,
            num_inference_steps=num_inference_steps,
            max_iter_to_alter=max_iter_to_alter,
            thresholds=thresholds,
            scale_factor=scale_factor,
        )
    else:
        out = sd_pipe(
            prompt=prompt,
            guidance_scale=guidance_scale,
            generator=generator,
            num_inference_steps=num_inference_steps,
        )
    return out.images[0]


def process_example(
    prompt: str,
    indices_to_alter_str: str,
    seed: int,
    apply_attend_and_excite: bool,
) -> tuple[list[tuple[int, str]], PIL.Image.Image]:
    token_table = get_token_table(prompt)
    result = run(
        prompt=prompt,
        indices_to_alter_str=indices_to_alter_str,
        seed=seed,
        apply_attend_and_excite=apply_attend_and_excite,
    )
    return token_table, result


with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(
        value="Duplicate Space for private use",
        elem_id="duplicate-button",
        visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
    )

    with gr.Row():
        with gr.Column():
            prompt = gr.Text(
                label="Prompt",
                max_lines=1,
                placeholder="A pod of dolphins leaping out of the water in an ocean with a ship on the background",
            )
            with gr.Accordion(label="Check token indices", open=False):
                show_token_indices_button = gr.Button("Show token indices")
                token_indices_table = gr.Dataframe(label="Token indices", headers=["Index", "Token"], col_count=2)
            token_indices_str = gr.Text(
                label="Token indices (a comma-separated list indices of the tokens you wish to alter)",
                max_lines=1,
                placeholder="4,16",
            )
            apply_attend_and_excite = gr.Checkbox(label="Apply Attend-and-Excite", value=True)
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            num_inference_steps = gr.Slider(
                label="Number of inference steps",
                minimum=1,
                maximum=MAX_INFERENCE_STEPS,
                step=1,
                value=50,
            )
            guidance_scale = gr.Slider(
                label="Guidance scale",
                minimum=0,
                maximum=50,
                step=0.1,
                value=7.5,
            )
            run_button = gr.Button("Generate")
        with gr.Column():
            result = gr.Image(label="Result")

    with gr.Row():
        examples = [
            [
                "A mouse and a red car",
                "2,6",
                2098,
                True,
            ],
            [
                "A mouse and a red car",
                "2,6",
                2098,
                False,
            ],
            [
                "A horse and a dog",
                "2,5",
                123,
                True,
            ],
            [
                "A horse and a dog",
                "2,5",
                123,
                False,
            ],
            [
                "A painting of an elephant with glasses",
                "5,7",
                123,
                True,
            ],
            [
                "A painting of an elephant with glasses",
                "5,7",
                123,
                False,
            ],
            [
                "A playful kitten chasing a butterfly in a wildflower meadow",
                "3,6,10",
                123,
                True,
            ],
            [
                "A playful kitten chasing a butterfly in a wildflower meadow",
                "3,6,10",
                123,
                False,
            ],
            [
                "A grizzly bear catching a salmon in a crystal clear river surrounded by a forest",
                "2,6,15",
                123,
                True,
            ],
            [
                "A grizzly bear catching a salmon in a crystal clear river surrounded by a forest",
                "2,6,15",
                123,
                False,
            ],
            [
                "A pod of dolphins leaping out of the water in an ocean with a ship on the background",
                "4,16",
                123,
                True,
            ],
            [
                "A pod of dolphins leaping out of the water in an ocean with a ship on the background",
                "4,16",
                123,
                False,
            ],
        ]
        gr.Examples(
            examples=examples,
            inputs=[
                prompt,
                token_indices_str,
                seed,
                apply_attend_and_excite,
            ],
            outputs=[
                token_indices_table,
                result,
            ],
            fn=process_example,
            cache_examples=os.getenv("CACHE_EXAMPLES") == "1",
            examples_per_page=20,
        )

    show_token_indices_button.click(
        fn=get_token_table,
        inputs=prompt,
        outputs=token_indices_table,
        queue=False,
        api_name="get-token-table",
    )

    gr.on(
        triggers=[prompt.submit, token_indices_str.submit, run_button.click],
        fn=randomize_seed_fn,
        inputs=[seed, randomize_seed],
        outputs=seed,
        queue=False,
        api_name=False,
    ).then(
        fn=get_token_table,
        inputs=prompt,
        outputs=token_indices_table,
        queue=False,
        api_name=False,
    ).then(
        fn=run,
        inputs=[
            prompt,
            token_indices_str,
            seed,
            apply_attend_and_excite,
            num_inference_steps,
            guidance_scale,
        ],
        outputs=result,
        api_name="run",
    )

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