app.py

import dataclasses

import gradio as gr
import spaces
import torch
from PIL import Image
from diffusers import DiffusionPipeline
from diffusers.utils import make_image_grid

DIFFUSERS_MODEL_IDS = [
    # SD Models
    "stabilityai/stable-diffusion-3-medium-diffusers",
    "stabilityai/stable-diffusion-xl-base-1.0",
    "stabilityai/stable-diffusion-2-1",
    "runwayml/stable-diffusion-v1-5",

    # Other Models
    "Prgckwb/trpfrog-diffusion",
]

EXTERNAL_MODEL_MAPPING = {
    "Beautiful Realistic Asians": "checkpoints/diffusers/Beautiful Realistic Asians v7",
}

MODEL_CHOICES = DIFFUSERS_MODEL_IDS + list(EXTERNAL_MODEL_MAPPING.keys())

current_model_id = "stabilityai/stable-diffusion-3-medium-diffusers"

device = "cuda" if torch.cuda.is_available() else "cpu"
if device == 'cuda':
    dtype = torch.float16
    pipe = DiffusionPipeline.from_pretrained(
        current_model_id,
        torch_dtype=dtype,
    )
    pipe.enable_sequential_cpu_offload()


@dataclasses.dataclass
class Input:
    prompt: str
    model_id: str = "stabilityai/stable-diffusion-3-medium-diffusers"
    negative_prompt: str = ''
    width: int = 1024
    height: int = 1024
    guidance_scale: float = 7.5
    num_inference_step: int = 28
    num_images: int = 4
    safety_checker: bool = True

    def to_list(self):
        return [
            self.prompt, self.model_id, self.negative_prompt,
            self.width, self.height, self.guidance_scale,
            self.num_inference_step, self.num_images, self.safety_checker
        ]


EXAMPLES = [
    Input(prompt='A cat holding a sign that says Hello world').to_list(),
    Input(
        prompt='Beautiful pixel art of a Wizard with hovering text "Achivement unlocked: Diffusion models can spell now"'
    ).to_list(),
    Input(prompt='A corgi wearing sunglasses says "U-Net is OVER!!"').to_list(),
    Input(
        prompt='Cinematic Photo of a beautiful korean fashion model bokeh train',
        model_id='Beautiful Realistic Asians',
        negative_prompt='(worst_quality:2.0) (MajicNegative_V2:0.8) BadNegAnatomyV1-neg bradhands cartoon, cgi, render, illustration, painting, drawing',
        width=512,
        height=512,
        guidance_scale=5.0,
        num_inference_step=50,
    ).to_list()
]


@spaces.GPU(duration=120)
@torch.inference_mode()
def inference(
        prompt: str,
        model_id: str = "stabilityai/stable-diffusion-3-medium-diffusers",
        negative_prompt: str = "",
        width: int = 512,
        height: int = 512,
        guidance_scale: float = 7.5,
        num_inference_steps: int = 50,
        num_images: int = 4,
        safety_checker: bool = True,
        progress=gr.Progress(track_tqdm=True),
) -> Image.Image:
    progress(0, "Starting inference...")

    global current_model_id, pipe

    progress(0.1, 'Loading pipeline...')
    if model_id != current_model_id:
        # For NOT Diffusers' Models
        if model_id not in DIFFUSERS_MODEL_IDS:
            model_id = EXTERNAL_MODEL_MAPPING[model_id]

        pipe = DiffusionPipeline.from_pretrained(
            model_id,
            torch_dtype=dtype,
        )
        pipe.enable_sequential_cpu_offload()

        current_model_id = model_id

    if not safety_checker:
        pipe.safety_checker = None

    if model_id not in DIFFUSERS_MODEL_IDS:
        progress(0.3, 'Loading Textual Inversion...')
        # Load Textual Inversion
        pipe.load_textual_inversion(
            "checkpoints/embeddings/BadNegAnatomyV1 neg.pt", token='BadNegAnatomyV1-neg'
        )

    # Generation
    progress(0.4, 'Generating images...')
    images = pipe(
        prompt,
        negative_prompt=negative_prompt,
        width=width,
        height=height,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        num_images_per_prompt=num_images,
    ).images

    if num_images % 2 == 1:
        image = make_image_grid(images, rows=num_images, cols=1)
    else:
        image = make_image_grid(images, rows=2, cols=num_images // 2)

    return image


if __name__ == "__main__":
    theme = gr.themes.Default(primary_hue=gr.themes.colors.emerald)

    with gr.Blocks(theme=theme) as demo:
        gr.Markdown(f"# Stable Diffusion Demo")

        with gr.Row():
            with gr.Column():
                prompt = gr.Text(label="Prompt", placeholder="Enter a prompt here")

                model_id = gr.Dropdown(
                    label="Model ID",
                    choices=MODEL_CHOICES,
                    value="stabilityai/stable-diffusion-3-medium-diffusers",
                )

                # Additional Input Settings
                with gr.Accordion("Additional Settings", open=False):
                    negative_prompt = gr.Text(label="Negative Prompt", value="", )

                    with gr.Row():
                        width = gr.Number(label="Width", value=512, step=64, minimum=64, maximum=2048)
                        height = gr.Number(label="Height", value=512, step=64, minimum=64, maximum=2048)
                        num_images = gr.Number(label="Num Images", value=4, minimum=1, maximum=10, step=1)

                    guidance_scale = gr.Slider(label="Guidance Scale", value=7.5, step=0.5, minimum=0, maximum=10)
                    num_inference_step = gr.Slider(
                        label="Num Inference Steps", value=50, minimum=1, maximum=100, step=2
                    )

                    with gr.Row():
                        safety_checker = gr.Checkbox(value=True, label='Use Safety Checker')

            with gr.Column():
                output_image = gr.Image(label="Image", type="pil")

        inputs = [
            prompt,
            model_id,
            negative_prompt,
            width,
            height,
            guidance_scale,
            num_inference_step,
            num_images,
            safety_checker
        ]

        btn = gr.Button("Generate")
        btn.click(
            fn=inference,
            inputs=inputs,
            outputs=output_image
        )

        gr.Examples(
            examples=EXAMPLES,
            inputs=inputs,
        )

    demo.queue().launch()