main.py

from fastapi import FastAPI, Response
from fastapi.middleware.cors import CORSMiddleware
import torch
from torch.cuda.amp import autocast
from diffusers import DiffusionPipeline
from diffusers import StableDiffusionPipeline
from io import BytesIO

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_credentials=True,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"]
)

# MODEL 1.4
model_id = "CompVis/stable-diffusion-v1-4"
device = "cuda" if torch.cuda.is_available() else "cpu"

pipe = StableDiffusionPipeline.from_pretrained(
    model_id, torch_dtype=torch.float32, cache_dir="./cache"
)

# MOdel 1.0
# model_id = "stabilityai/stable-diffusion-xl-base-1.0"
# device = "cuda" if torch.cuda.is_available() else "cpu"
# pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32, use_safetensors=True, variant="fp16")

pipe = pipe.to(device)


@app.get("/")
def generate(prompt: str):
    with autocast(device):
        image = pipe(prompt, guidance_scale=8.5).images[0]

    buffer = BytesIO()
    image.save(buffer, format="PNG")
    buffer.seek(0)

    return Response(content=buffer.getvalue(), media_type="image/png")