main.py

from fastapi import FastAPI
import gradio as gr
import torch
from sd import pipeline
from sd import model_loader
from transformers import AutoTokenizer
from diffusers import StableDiffusionPipeline

app = FastAPI()

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

tokenizer = AutoTokenizer.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="tokenizer")
weights_url = "https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.ckpt"
models = model_loader.from_pretrained(weights_url, device)

pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", use_safetensors=True)
pipe = pipe.to(device)

MIN_IMAGE_SIZE = 256
MAX_IMAGE_SIZE = 1024

MAX_SEED = 2147483647 # 2^31 - 1

def generate_image(prompt, negative_prompt, seed, randomize_seed, guidance_scale, num_inference_steps, model, width, height):

    if randomize_seed:
        seed = torch.randint(0, MAX_SEED, (1,)).item()
        
    generator = torch.Generator(device=device).manual_seed(seed)

    if model == "from-scratch":
        image = pipeline.generate(
            prompt=prompt,
            uncond_prompt=negative_prompt,
            input_image=None,
            strength=0.9,
            cfg_scale=guidance_scale,
            n_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
            device=device,
            idle_device="cpu",
            models=models,
            tokenizer=tokenizer,
        )
    else:
        image = pipe(
            prompt=prompt,
            negative_prompt=negative_prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
        ).images[0]

    return image

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

md = """
# Text-to-Image: Stable Diffusion from Scratch
### By [Nazareno Amidolare](https://kepler296e.github.io/)
Using **Docker**, **FastAPI**, **PyTorch** and **Gradio**.
### References
- [Coding Stable Diffusion from scratch in PyTorch](https://www.youtube.com/watch?v=ZBKpAp_6TGI&ab_channel=UmarJamil)
- [Hugging Space Diffusers](https://github.com/huggingface/diffusers/)

Currently running on a **CPU** (≈20 minutes per image).
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(md)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            run_button = gr.Button("Run", scale=0)
        
        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):

            with gr.Row():

                model = gr.Dropdown(
                    label="Model",
                    choices=["from-scratch", "runwayml/stable-diffusion-v1-5"],
                    value="from-scratch",
                    interactive=True,
                )

            with gr.Row():

                seed = gr.Slider(
                    label="Seed",
                    minimum=0,
                    maximum=MAX_SEED,
                    step=1,
                    value=42,
                )

                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)

            with gr.Row():
            
                negative_prompt = gr.Text(
                    label="Negative prompt",
                    max_lines=1,
                    placeholder="Enter a negative prompt",
                    visible=False,
                    value="",
                )
                        
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=MIN_IMAGE_SIZE,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=MIN_IMAGE_SIZE,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=512,
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=1.0,
                    maximum=14.0,
                    step=0.1,
                    value=8.0,
                )
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=50,
                )

    run_button.click(
        fn = generate_image,
        inputs = [prompt, negative_prompt, seed, randomize_seed, guidance_scale, num_inference_steps, model, width, height],
        outputs = [result]
    )

app = gr.mount_gradio_app(app, demo, "/")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)