sanford

prompt = """我想让你对小说内容进行分镜，根据原文描述推断出的场景；推断和补充缺失或隐含的信息，包括但不限于：人物衣服，人物发型，人物发色，人物脸色，人物五官特点，人物体态，人物情绪，人物肢体动作等）、风格描述（包括但不限于：年代描述、空间描述、时间段描述、地理环境描述、天气描述）、物品描述（包括但不限于：动物、植物、食物、水果、玩具）、画面视角（包括但不限于：人物比例、镜头深度描述、观察角度描述），但不要过度。通过镜头语言描述，描绘更丰富的人物情绪和情感状态，你理解后通过句子生成一段新的描述内容。输出格式改为：插画一：原文描述：对应的原文全部句子；画面描述：对应的画面剧情内容；画面角色：画面中出现的角色名称；穿着：主角穿着便装；位置：坐在吧台前；表情：面部线条温和，表情惬意； 行为：手上轻轻晃动着手中的酒杯。环境：吧台的背景是暗调的，烛光在背景中摇曳，给人一种迷离的感觉。如果你理解了这一点要求，请确认这五点要求，返回结果只要这五点的内容，小说内容如下:"""

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')