电影级 Kling

通过组合以下三要素，生成一段 5 秒钟的电影级视频：

• 一个 角色（人物/模特）
• 一件 物品（产品、物体、道具）
• 一处 地点（背景、环境）

系统内部会先分别生成角色设定表（Character Sheet）和物品设定表（Item Sheet），再将二者输入 Kling 进行视频合成。

部署信息

• 部署 ID： e5258667-dec2-438f-84d0-f22049692483
• API 地址： POST https://api.comfydeploy.com/api/run/deployment/queue
• 认证方式： Authorization: Bearer $COMFY_DEPLOY_API_KEY

输入参数

角色设定表提示词结构

该提示词分为 两部分 —— 仅 [brackets] 部分可变：

Create a photorealistic character sheet. Include a closeup portrait which is left aligned (the outfit must be visible in the portrait), no borders, with a neutral expression and then also include a full view shot that shows the front, right side view, left side view and back of the character. The character is placed on a white background. Don't include any text. No borders, no soft gradients. [CLOTHING DESCRIPTION HERE]

规则： 将 [CLOTHING DESCRIPTION HERE] 替换为符合当前场景的服装描述；方括号前的所有内容必须严格保持原样不变。

示例：

• [They are wearing a tailored black suit with a white shirt, no tie, luxury editorial style.]
• [They are wearing dirty damaged clothes in the style of The Last Of Us, dystopic aesthetic.]
• [They are wearing a red Nike tracksuit and white sneakers, athletic style.]

视频提示词编写指南

• 保持 简洁 —— 描述动作/交互即可，无需面面俱到
• 通过动作自然隐含 角色 + 物品 + 地点 三要素
• 可在 5 秒内建议 1–2 次镜头切换（cut）
• 适用场景包括：广告、短片、音乐视频、产品演示、动作片段等

优质示例：

• "Short cinematic sequence of the character dunking a basketball in the ring while wearing the green Shaq shoes."
• "Character walks into the desert location, picks up the item from the sand, holds it toward camera. Quick cut to close-up of the item."
• "Product reveal: character turns around in the urban location, tosses the item to camera in slow motion."

不良示例（过于详细）：

• ❌ "Camera starts at f/1.8 aperture, golden hour light at 47 degrees, character moves exactly 3 steps forward..."

第 0 步 — 核验各图像身份（强制步骤）

务必使用 image 工具预先识别每张图像类型，严禁按文件顺序主观假设。

image(images=[file1, file2, file3], prompt="Para cada imagen, decí brevemente si es: persona, producto/objeto, o lugar/ambiente")

随后按如下规则分配：

• character_base = 人物/模特
• item_base = 其所交互的 产品/物体
• location_base = 场所/环境/背景

第 1 步 — 将图像上传至 ComfyDeploy 存储

务必先将本地文件上传至 ComfyDeploy 自有存储空间。禁止使用外部或自定义域名。

source ~/clawd/.env

upload_to_comfy() {
  curl -s -X POST "https://api.comfydeploy.com/api/file/upload" 
    -H "Authorization: Bearer $COMFY_DEPLOY_API_KEY" 
    -F "file=@$1" | jq -r '.file_url'
}

URL_CHAR=$(upload_to_comfy /path/to/character.jpg)
URL_ITEM=$(upload_to_comfy /path/to/item.jpg)
URL_LOC=$(upload_to_comfy /path/to/location.jpg)
# 返回示例：https://comfy-deploy-output.s3.us-east-2.amazonaws.com/inputs/img_XXXX.jpg

第 2 步 — 提交运行任务

RUN_ID=$(curl -s -X POST "https://api.comfydeploy.com/api/run/deployment/queue" 
  -H "Content-Type: application/json" 
  -H "Authorization: Bearer $COMFY_DEPLOY_API_KEY" 
  -d "{
    "deployment_id": "e5258667-dec2-438f-84d0-f22049692483",
    "inputs": {
      "character_base": "$URL_CHAR",
      "item_base": "$URL_ITEM",
      "location_base": "$URL_LOC",
      "character_sheet_prompt": "Create a photorealistic character sheet. Include a closeup portrait which is left aligned (the outfit must be visible in the portrait), no borders, with a neutral expression and then also include a full view shot that shows the front, right side view, left side view and back of the character. The character is placed on a white background. Don't include any text. No borders, no soft gradients. [CLOTHING DESCRIPTION]",
      "video_prompt": "VIDEO_SCENE_DESCRIPTION",
      "input_seed": -1
    }
  }" | jq -r '.run_id')

echo "Run ID: $RUN_ID"

轮询获取结果

while true; do
  RESULT=$(curl -s "https://api.comfydeploy.com/api/run/$RUN_ID" 
    -H "Authorization: Bearer $COMFY_DEPLOY_API_KEY")
  STATUS=$(echo $RESULT | jq -r '.status')
  echo "Status: $STATUS"
  if [ "$STATUS" = "success" ]; then
    echo $RESULT | jq '.outputs'
    break
  elif [ "$STATUS" = "failed" ]; then
    echo "Failed!"
    break
  fi
  sleep 10
done

第 3 步 — 轮询并下载结果

# 持续轮询直至成功
while true; do
  RESULT=$(curl -s "https://api.comfydeploy.com/api/run/$RUN_ID" 
    -H "Authorization: Bearer $COMFY_DEPLOY_API_KEY")
  STATUS=$(echo $RESULT | jq -r '.status')
  [ "$STATUS" = "success" ] && break
  [ "$STATUS" = "failed" ] && echo "FAILED" && break
  sleep 15
done

# 下载输出至允许目录
mkdir -p ~/clawd/output/cinematic-kling
BASE_URL="https://comfy-deploy-output.s3.us-east-2.amazonaws.com/outputs/runs/$RUN_ID"
curl -sL "$BASE_URL/CS_2_00001_.png" -o ~/clawd/output/cinematic-kling/character-sheet.png
curl -sL "$BASE_URL/ITEM_00001_.png" -o ~/clawd/output/cinematic-kling/item-sheet.png
curl -sL "$BASE_URL/LOCATION_00001_.png" -o ~/clawd/output/cinematic-kling/location-sheet.png
curl -sL "$BASE_URL/ComfyUI_00001_.mp4" -o ~/clawd/output/cinematic-kling/video.mp4

第 4 步 — 以文件形式发送结果

务必通过 filePath 发送文件附件 — 禁止直接发送原始 URL。

message(action=send, channel=telegram, target=USER_ID, filePath=~/clawd/output/cinematic-kling/character-sheet.png, message="Character Sheet")
message(action=send, channel=telegram, target=USER_ID, filePath=~/clawd/output/cinematic-kling/item-sheet.png, message="Item Sheet")
message(action=send, channel=telegram, target=USER_ID, filePath=~/clawd/output/cinematic-kling/location-sheet.png, message="Location Sheet")
message(action=send, channel=telegram, target=USER_ID, filePath=~/clawd/output/cinematic-kling/video.mp4, message="🎬 Video")

输出结构

Python 辅助函数

import requests, os, time

def cinematic_kling(character_url, item_url, location_url, clothing_desc, video_prompt, seed=-1):
    """
    character_url: 角色图像的公开 URL
    item_url: 物品图像的公开 URL
    location_url: 地点图像的公开 URL
    clothing_desc: 角色设定表提示词中 [brackets] 内的服装描述
    video_prompt: 简洁的场景描述
    返回值: 包含 'images' 列表与 'video' URL 的字典
    """
    API_KEY = os.environ['COMFY_DEPLOY_API_KEY']
    
    character_sheet_prompt = (
        "Create a photorealistic character sheet. Include a closeup portrait which is left aligned "
        "(the outfit must be visible in the portrait), no borders, with a neutral expression and then "
        "also include a full view shot that shows the front, right side view, left side view and back "
        "of the character. The character is placed on a white background. Don't include any text. "
        f"No borders, no soft gradients. [{clothing_desc}]"
    )
    
    # 提交任务
    r = requests.post(
        "https://api.comfydeploy.com/api/run/deployment/queue",
        headers={"Content-Type": "application/json", "Authorization": f"Bearer {API_KEY}"},
        json={
            "deployment_id": "e5258667-dec2-438f-84d0-f22049692483",
            "inputs": {
                "character_base": character_url,
                "item_base": item_url,
                "location_base": location_url,
                "character_sheet_prompt": character_sheet_prompt,
                "video_prompt": video_prompt,
                "input_seed": seed
            }
        }
    )
    run_id = r.json()['run_id']
    print(f"Run ID: {run_id}")
    
    # 轮询状态
    while True:
        result = requests.get(
            f"https://api.comfydeploy.com/api/run/{run_id}",
            headers={"Authorization": f"Bearer {API_KEY}"}
        ).json()
        status = result.get('status')
        print(f"Status: {status}")
        if status == 'success':
            return result.get('outputs', {})
        elif status == 'failed':
            raise Exception(f"Run failed: {result}")
        time.sleep(10)

典型运行耗时

约 3–5 分钟（含角色设定表生成 + Kling 视频生成）

典型应用场景

• 产品广告： 角色在品牌化场景中使用/穿戴该物品
• 短片制作： 角色在电影级环境中与道具互动
• 音乐视频： 艺术家 + 乐器/道具 + 舞台/场景
• 时尚展示： 模特在特定环境中穿着服饰单品
• 游戏/废土风格： 类《最后生还者》风格的角色，在末日场景中手持游戏内物品

注意事项

• character_sheet_prompt 的固定部分严禁修改 —— 仅允许调整 [brackets] 中的内容
• input_seed: -1 表示每次运行使用随机种子；设置固定整数可确保结果可复现
• 视频长度固定为 5 秒
• 所有输出均来自 ComfyDeploy 内置 S3 存储 —— 无需手动上传