app.py

import gradio as gr
import spaces
import torch
from PIL import Image
from diffusers import DiffusionPipeline

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

EXTERNAL_MODEL_URL_MAPPING = {
    "Beautiful Realistic Asians": "https://civitai.com/api/download/models/177164?type=Model&format=SafeTensor&size=full&fp=fp16",
}

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

# Global Variables
current_model_id = "stabilityai/stable-diffusion-3-medium-diffusers"
device = "cuda" if torch.cuda.is_available() else "cpu"

if device == 'cuda':
    pipe = DiffusionPipeline.from_pretrained(
        current_model_id,
        torch_dtype=torch.float16,
    ).to(device)



@spaces.GPU()
@torch.inference_mode()
def inference(
        model_id: str,
        prompt: str,
        negative_prompt: str = "",
        progress=gr.Progress(track_tqdm=True),
) -> Image.Image:
    progress(0, "Starting inference...")

    global current_model_id, pipe

    if model_id != current_model_id:
        try:
            pipe = DiffusionPipeline.from_pretrained(
                model_id,
                torch_dtype=torch.float16,
            ).to(device)
            current_model_id = model_id
        except Exception as e:
            raise gr.Error(str(e))

    image = pipe(
        prompt,
        negative_prompt=negative_prompt,
    ).images[0]

    return image


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

        with gr.Row():
            with gr.Column():
                inputs = [
                    gr.Dropdown(
                        label="Model ID",
                        choices=MODEL_CHOICES,
                        value="stabilityai/stable-diffusion-3-medium-diffusers",
                    ),
                    gr.Text(label="Prompt", value=""),
                    gr.Text(label="Negative Prompt", value=""),
                ]

                with gr.Accordion("Additional Settings (W.I.P)", open=False):
                    with gr.Row():
                        width_component = gr.Number(label="Width", value=512, step=64, minimum=64, maximum=1024)
                        height_component = gr.Number(label="Height", value=512, step=64, minimum=64, maximum=1024)

                    additional_inputs = [
                        width_component,
                        height_component,
                        gr.Number(label="Guidance Scale", value=7.5, step=0.5, minimum=0, maximum=10),
                        gr.Slider(label="Num Inference Steps", value=None, minimum=1, maximum=1000, step=1)
                    ]

            with gr.Column():
                outputs = [
                    gr.Image(label="Image", type="pil"),
                ]

        gr.Examples(
            examples=[
                ['stabilityai/stable-diffusion-3-medium-diffusers', 'A cat holding a sign that says Hello world', ""],
                ['stabilityai/stable-diffusion-3-medium-diffusers', 'Beautiful pixel art of a Wizard with hovering text "Achivement unlocked: Diffusion models can spell now"', ''],
                ['stabilityai/stable-diffusion-3-medium-diffusers', 'A corgi wearing sunglasses says "U-Net is OVER!!"', ''],
            ],
            inputs=inputs,
        )

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

    demo.queue().launch()