app.py

"""See https://huggingface.co/spaces/Gradio-Blocks/Story-to-video/blob/main/app.py."""
import base64
import io
import logzero
import os
import re
import time
from random import choice, choices

import gradio as gr
import translators as ts
from fastlid import fastlid
from logzero import logger
from PIL import Image  # opencv-python
from tenacity import retry
from tenacity.stop import stop_after_attempt, stop_after_delay

# from PIL import Image
# from transformers import AutoTokenizer, AutoModelForSeq2SeqLM,pipeline
# import requests
# import torch

image_gen = gr.Interface.load("spaces/multimodalart/latentdiffusion")
# image_gen = gr.Interface.load("huggingface/multimodalart/latentdiffusion")

os.environ["TZ"] = "Asia/Shanghai"
try:
    time.tzset()
except Exception:
    ...  # Windows

loglevel = 10  # change to 20 to switch off debug
logzero.loglevel(loglevel)


examples_ = [
    "黄金在河里流淌，宝石遍地，空中铺满巨大的彩虹。",
    "蓝色的夜，森林中好多萤火虫",
    "黑云压城城欲摧 ，甲光向日金鳞开。",
    "季姬寂，集鸡，鸡即棘鸡。棘鸡饥叽，季姬及箕稷济鸡。",
    "an apple",
    "a cat",
    "blue moon",
    "metaverse",
]


# @retry(stop=stop_after_attempt(5))
@retry(stop=(stop_after_delay(10) | stop_after_attempt(5)))
def tr_(
    text: str,
    from_language="auto",
    to_language="en",
) -> str:
    """Wrap [ts.deepl, ts.baidu, ts.google] with tenacity.

    not working: sogou;  ?tencent
    """
    cand = [ts.baidu, ts.youdao, ts.google]
    for tr in [ts.deepl] + choices(cand, k=len(cand)):
        try:
            res = tr(
                text,
                from_language=from_language,
                to_language=to_language,
            )
            logger.info(" api used: %s", tr.__name__)
            tr_.api_used = tr.__name__
            break
        except Exception:
            continue
    else:
        res = "Something is probably wrong, ping dev to fix it if you like."

    return res


def generate_images(phrase: str, steps: int = 125):
    if not phrase.strip():
        phrase = choice(examples_)

    generated_text = phrase
    detected = "en"
    extra_info = ""
    try:
        detected = fastlid(phrase)[0]
    except Exception as exc:
        logger.error(exc)

    logzero.loglevel(loglevel)
    logger.debug("phrase: %s, deteted: %s", phrase, detected)

    # safeguard short Chinese phrases
    if len(phrase) <= 10 and re.search(r"[一-龟]+", phrase):
        detected = "zh"
        logger.debug(" safeguard branch ")

    if detected not in ["en"]:
        try:
            generated_text = tr_(
                phrase,
                detected,
                "en",
            )
            extra_info = f"({tr_.api_used}: {generated_text})"
        except Exception as exc:
            logger.error(exc)
            return None, f"{phrase:}, errors: {str(exc)}"

    # steps = 125
    width = 256
    height = 256
    num_images = 4
    num_images = 1
    diversity = 6

    try:
        image_bytes = image_gen(
            generated_text, steps, width, height, num_images, diversity
        )
    except Exception as exc:
        logger.error(exc)
        return None, f"phrase: {phrase}, errors: {str(exc)}. Try again."

    # Algo from spaces/Gradio-Blocks/latent_gpt2_story/blob/main/app.py
    # generated_images = []

    img = None
    err_msg = f"{phrase} {extra_info}"
    for image in image_bytes[1]:
        image_str = image[0]
        try:
            image_str = image_str.replace("data:image/png;base64,", "")
        except Exception as exc:
            logger.error(exc)
            err_msg = str(exc)
            return None, f"errors: {err_msg}. Try again."
        decoded_bytes = base64.decodebytes(bytes(image_str, "utf-8"))
        img = Image.open(io.BytesIO(decoded_bytes))

        # generated_images.append(img)

    # return generated_images

    return img, err_msg


# examples = [["an apple", 125], ["Donald Trump", 125]]
examples = [list(_) for _ in zip(examples_, [125] * len(examples_))]

inputs = [
    # "text",
    gr.Text(value="a dog with a funny hat"),
    gr.Slider(minimum=2, maximum=250, value=115, step=5),
]

iface = gr.Interface(
    generate_images,
    inputs,
    # ["image", gr.Text(value="", label="phrase")],
    [gr.Image(label=""), gr.Text(value="", label="phrase")],
    examples=examples,
    cache_examples=False,
    allow_flagging="never",
)

iface.launch(enable_queue=True)