AI 内容变现五大赛道工程化：从漫剧到数字人的统一生产管线

from dataclasses import dataclass, field

@dataclass
class Asset:
    id: str
    kind: str          # character / scene / motion / voice
    content: dict      # 结构化描述，不是原始文件
    ref_image: str = None
    locked: bool = False   # 一旦锁定，后续生成必须对齐

@dataclass
class Project:
    topic: str
    assets: dict = field(default_factory=dict)
    shots: list = field(default_factory=list)
    output_type: str = "comic"  # comic / oral / dance / painting

import json
from openai import OpenAI

client = OpenAI(api_key="sk-xxx", base_url="https://api.deepseek.com/v1")

def define_character(name: str, style: str) -> Asset:
    """生成人物设定，产出一段结构化描述"""
    prompt = f"""为角色「{name}」生成视觉设定。
风格：{style}
输出 JSON：{{"appearance":str,"outfit":str,"color_palette":str,
"personality":str}}"""
    resp = client.chat.completions.create(
        model="deepseek-chat",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.7,
        response_format={"type": "json_object"})
    return Asset(id=name, kind="character",
                 content=json.loads(resp.choices[0].message.content))

def render_asset(asset: Asset, seed: int = 42) -> str:
    """根据锁定设定出图，固定 seed 保证一致性"""
    c = asset.content
    prompt = f"{c['appearance']}, {c['outfit']}, {c['color_palette']}"
    resp = client.images.generate(
        model="seedream", prompt=prompt,
        size="1024x1024", seed=seed)
    asset.ref_image = resp.data[0].url
    asset.locked = True
    return asset.ref_image

def check_consistency(asset: Asset, new_image: str) -> dict:
    """对比新图与参考图，判断是否同一角色"""
    resp = client.chat.completions.create(
        model="deepseek-vl",
        messages=[{"role": "user", "content": [
            {"type": "text", "text": "这两张图是同一个人物吗？"
             "输出 JSON：{\"same\":bool,\"diff\":str}"},
            {"type": "image_url", "image_url": {"url": asset.ref_image}},
            {"type": "image_url", "image_url": {"url": new_image}}
        ]}],
        response_format={"type": "json_object"})
    return json.loads(resp.choices[0].message.content)

def produce(project: Project) -> dict:
    if project.output_type == "comic":
        return _render_comic(project)       # 分镜图 → 拼接 → 对白
    elif project.output_type == "oral":
        return _render_oral(project)        # 数字人 → 音频 → 口型对齐
    elif project.output_type == "dance":
        return _render_dance(project)       # 动作模板 → 骨骼绑定 → 视频
    elif project.output_type == "painting":
        return _render_painting(project)    # 系列出图 → 筛选 → 打包
    elif project.output_type == "beast":
        return _render_beast(project)       # 异兽设定 → 场景合成 → 短视频

def _render_oral(project: Project) -> dict:
    """数字人口播：音频驱动口型"""
    host = project.assets["host"]
    script = project.shots[0]["text"]
    audio = client.audio.speech.create(
        model="tts", voice=host.content["voice"], input=script)
    video = client.video.generate(
        model="digital-human",
        image_url=host.ref_image,
        audio=audio.content,
        prompt="自然口播，轻微手势")
    return {"video": video.data[0]["url"]}