sanford

prompt = """我想讓你對小說內容進行分鏡，根據原文描述推斷出的場景；推斷和補充缺失或隱含的信息，包括但不限於：人物衣服，人物髮型，人物髮色，人物臉色，人物五官特點，人物體態，人物情緒，人物肢體動作等）、風格描述（包括但不限於：年代描述、空間描述、時間段描述、地理環境描述、天氣描述）、物品描述（包括但不限於：動物、植物、食物、水果、玩具）、畫面視角（包括但不限於：人物比例、鏡頭深度描述、觀察角度描述），但不要過度。通過鏡頭語言描述，描繪更豐富的人物情緒和情感狀態，你理解後通過句子生成一段新的描述內容。輸出格式改為：插畫一：原文描述：對應的原文全部句子；畫面描述：對應的畫面劇情內容；畫面角色：畫面中出現的角色名稱；穿著：主角穿著便裝；位置：坐在吧台前；表情：面部線條溫和，表情愜意； 行為：手上輕輕晃動著手中的酒杯。環境：吧台的背景是暗調的，燭光在背景中搖曳，給人一種迷離的感覺。如果你理解了這一點要求，請確認這五點要求，返回結果只要這五點的內容，小說內容如下:"""

def split_text_into_chunks(text, max_length=ai_max_length):
    """
    Split text into chunks with a maximum length, ensuring that splits only occur at line breaks.
    """
    lines = text.splitlines()
    chunks = []
    current_chunk = ''
    for line in lines:
        if len(current_chunk + ' ' + line) <= max_length:
            current_chunk += ' ' + line
        else:
            chunks.append(current_chunk)
            current_chunk = line
    chunks.append(current_chunk)
    return chunks

def rewrite_text_with_genai(text, prompt="Please rewrite this text:"):
    chunks = split_text_into_chunks(text)
    rewritten_text = ''
    # pbar = tqdm(total=len(chunks), ncols=150)
    genai.configure(api_key=cfg['genai_api_key'])
    model = genai.GenerativeModel('gemini-pro')
    for chunk in chunks:
        _prompt=f"{prompt}\n{chunk}",
        response = model.generate_content(
            contents=_prompt, 
            generation_config=genai.GenerationConfig(
                temperature=0.1,
            ),
            stream=True,
            safety_settings = [
                {
                    "category": "HARM_CATEGORY_DANGEROUS",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_HARASSMENT",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_HATE_SPEECH",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
                    "threshold": "BLOCK_NONE",
                },
            ]
        )
        for _chunk in response:
            if _chunk.text is not None:
                rewritten_text += _chunk.text.strip()
    #     pbar.update(1)
    # pbar.close()
    return rewritten_text

from diffusers import StableDiffusionPipeline
from diffusers.utils import load_image
import torch



model_path = "./models/cetusMix_Whalefall2.safetensors"
pipeline = StableDiffusionPipeline.from_single_file(
    model_path,
    torch_dtype=torch.float16,
    variant="fp16"
    ).to("mps")
generator = torch.Generator("mps").manual_seed(31)

def sd_cetus(save_name, prompt):
    prompt = prompt
    image = pipeline(prompt).images[0]
    image.save('data/img/'+ save_name +'.jpg')

import edge_tts
import asyncio



voice = 'zh-CN-YunxiNeural'
output = 'data/voice/'
rate = '-4%'
volume = '+0%'

async def tts_function(text, save_name):
    tts = edge_tts.Communicate(
        text,
        voice=voice,
        rate=rate,
        volume=volume
        )
    await tts.save(output + save_name + '.wav')