main.py

import os
import time
import sys
import re
import random
import torch
from PIL import Image
from fastapi import FastAPI
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel
from pydantic import Field
from diffusers import StableDiffusionPipeline
from transformers import (
    pipeline,
    MBart50TokenizerFast,
    MBartForConditionalGeneration,
)

app = FastAPI()

class Data(BaseModel):
	string: str
	member_secret: str

class ItemOut(BaseModel):
	status: str
	file: str

@app.get("/")
def index():
	return "SORRY! This file is member only."

@app.post("/draw", response_model=ItemOut)
def draw(data: Data):
	if data.member_secret != "" and data.member_secret == os.environ.get("MEMBER_SECRET"):

		device = "cuda" if torch.cuda.is_available() else "cpu"
		device_dict = {"cuda": 0, "cpu": -1}

		seedno = 0
		if '_seed' in data.string:
			seed = 1024
			stext = re.search(r'_seed[1-4]?', data.string)
			stext = re.sub('_seed', '', stext.group())
			if stext.isnumeric() and 0 < int(stext) < 5:
				seedno = stext
		else:
		 	seed = random.randrange(1024)

		text = re.sub('^#', '', data.string)
		text = re.sub('_seed[1-4]?', '', text)

		# prompt = '(('+text+')) (( photograph )), highly detailed, sharp focus, 8k, 4k, (( photorealism )), detailed, saturated, portrait, 50mm, F/2.8, 1m away, ( global illumination, studio light, volumetric light ), ((( multicolor lights )))'
		prompt = '(('+text+')) (( photograph )), highly detailed, sharp focus, 8k, 4k, (( photorealism )), detailed, saturated, portrait, 50mm, F/2.8, 1m away, ((( multicolor lights )))'
		n_prompt = 'text, blurry, art, painting, rendering, drawing, sketch, (( ugly )), (( duplicate )), ( morbid ), (( mutilated )), ( mutated ), ( deformed ), ( disfigured ), ( extra limbs ), ( malformed limbs ), ( missing arms ), ( missing legs ), ( extra arms ), ( extra legs ), ( fused fingers ), ( too many fingers ), long neck, low quality, worst quality'

		# https://huggingface.co/docs/hub/spaces-sdks-docker-first-demo
		# how to validation: https://qiita.com/bee2/items/75d9c0d7ba20e7a4a0e9
		# https://github.com/huggingface/diffusers
		# https://github.com/huggingface/diffusers/pull/1142

		# Add language detection pipeline
		language_detection_model_ckpt = "papluca/xlm-roberta-base-language-detection"
		language_detection_pipeline = pipeline("text-classification",
		                                       model=language_detection_model_ckpt,
		                                       device=device_dict[device])

		# Add model for language translation
		trans_tokenizer = MBart50TokenizerFast.from_pretrained("facebook/mbart-large-50-many-to-one-mmt")
		trans_model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-one-mmt").to(device)

		model_id = 'stabilityai/stable-diffusion-2'

		#pipe = StableDiffusionPipeline.from_pretrained(model_id)
		pipe = StableDiffusionPipeline.from_pretrained(
			model_id, 
			custom_pipeline="multilingual_stable_diffusion",
			detection_pipeline=language_detection_pipeline,
			translation_model=trans_model,
			translation_tokenizer=trans_tokenizer,
			revision='fp16', 
			torch_dtype=torch.float16
		)
		pipe.enable_attention_slicing()	# reduce gpu usage
		pipe = pipe.to(device)

		generator = torch.Generator("cuda").manual_seed(seed)
		images = pipe(prompt, negative_prompt=n_prompt, guidance_scale=7.5, generator=generator, num_images_per_prompt=1).images

		if 0 < int(seedno) < 5:
			grid = images[(int(seedno) - 1)]
		else:
			# Limit of T4 small...
			grid = image_grid(images, rows=1, cols=1)

		fileName = "sd_" + str(time.time()) + '.png'
		grid.save("/code/tmpdir/" + fileName)

		print(fileName)

		return {"status": "OK", "file": fileName}
	else:
		return {"status": "SORRY! This file is member only.", "file": ""}

app.mount("/static", StaticFiles(directory="/code/tmpdir"), name="/static")

# helper function taken from: https://huggingface.co/blog/stable_diffusion
def image_grid(imgs, rows, cols):
    assert len(imgs) == rows*cols

    w, h = imgs[0].size
    grid = Image.new('RGB', size=(cols*w, rows*h))
    grid_w, grid_h = grid.size

    for i, img in enumerate(imgs):
        grid.paste(img, box=(i%cols*w, i//cols*h))
    return grid