SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsaigc-director

aigc-director

Agents & models

View repository

Updated 10-13 15:56

About this skill

AIGC-Director Agent Skill

Run locally: This is alocally deployedvideo generation project,both frontend and backend run on this machine：

  • Backend:http://localhost:8000
  • Frontend:http://localhost:3000
  • All API calls should request the local server; do not request other addresses!
  • Make sure that before calling any API, both the backend and frontend services have been started and are running properly!

Core idea: The Agent should be like a "continuously accompanying intelligent video production assistant". Each time it completes an important user-perceivable task, it should immediately give the user a brief report and wait for user confirmation.

Core principle: The output of each stage must be shown to the user, and it must stop and wait for user confirmation before continuing to the next stage.

Prevent forgetting: Throughout the entire workflow, the Agent may forget previous user input or the output content of previous stages.Whenever entering a new stage, the Agent must reload this SKILL document to ensure it does not forget any details。

Project structure

aigc-director/                    ← OpenClaw 调用的 skill 根目录
├── aigc-claw/                    ← 前后端项目代码
│   ├── backend/                  ← FastAPI 后端（端口 8000）
│   │   └── code/result/          ← 模型生成产物存放目录
│   │            ├── script/      ← 剧本产物
│   │            ├── image/       ← 图片产物（角色、场景、参考图）
│   │            └── video/       ← 视频产物
│   └── frontend/                 ← Next.js 前端（端口 3000）
├── references/                   ← OpenClaw 调用时的参考文档
│   ├── init_project/             ← 项目初始化
│   ├── run_project/              ← 服务启动
│   ├── workflow/                 ← 六阶段工作流 API
│   ├── sandbox/                  ← 临时工作台 API
│   └── send_message/             ← 消息发送
└── SKILL.md                      ← skill 正文

Output storage directory：aigc-claw/backend/code/result/

  • script/ - Script output
  • image/ - Image output (characters, scenes, reference images)
  • video/ - Video output

Stages and stop points (9 total)

Stop point Stage phase value Description Action
1 Project configuration - Confirm configuration options Show configuration → user confirmation
2 Script generation suggest_expand Suggest expansion Wait for user confirmation
3 Script generation logline_selection Choose plot Choose from 3 candidates
4 Script generation mode_selection Choose mode Movie (4 acts) / Short film (1 act)
5 Script generation script_generation Confirm script Continue after confirmation
6 Character/scene design - Confirm character/scene images Continue after confirmation
7 Storyboard design - Confirm storyboard list Continue after confirmation
8 Reference image generation - Confirm reference images Continue after confirmation
9 Video generation - Confirm video clips Continue after confirmation

Note：suggest_expand and logline_selection may be skipped based on input quality.

Workflow

1. Local deployment (executed only during initialization)

When the user requests "initialize project", "configure project", or "deploy project", project initialization must be performed first: refer to init_all.md to execute the complete initialization workflow.

Note: Use this only when the user downloads the project for the first time or needs to reconfigure the environment. If the project has already been initialized, skip this step and directly check the service running status.

2. Check local services

Refer to start_backend.md and start_frontend.md to check whether the service is running.

⚠️ Mandatory requirement: If the service is not running, you must start the service before continuing!

2. Routing decision

The user says process
"generate image" temporary workspace (sandbox)
"generate video" you must first ask: long video (workflow) or short video (workspace)?
"analyze image" temporary workspace (sandbox)
"ask LLM question" temporary workspace (sandbox)
"photo to anime" temporary workspace (sandbox)

3. Execution flow

1. 检查后端运行状态 → 未运行则参考 start_backend.md 启动 → 等待3秒 → 再次检查
2. 检查前端运行状态 → 未运行则参考 start_frontend.md 启动 → 等待5秒 → 再次检查
3. 检查 API Key 配置 → 读取 .env 文件，确认所需 API Key 已配置
4. 参考 create_project.md 询问用户项目配置 → 停点1（配置确认）→ 创建项目
5. 参考 create_script.md 执行剧本生成 → 停点2-5
6. 参考 create_character.md 执行角色设计 → 停点6
7. 参考 create_storyboard.md 执行分镜设计 → 停点7
8. 参考 create_reference.md 执行参考图生成 → 停点8
9. 参考 create_video.md 执行视频生成 → 停点9
10. 参考 create_post.md 执行后期剪辑
11. 完成 → 发送最终视频给用户

Note: Be sure to refer to the specific documents under the references/ directory to execute each step. Do not call APIs from memory or assumptions!

Check API Key configuration

Before creating a project, you must check whether the API Key corresponding to the model selected by the user has been configured:

# 读取 .env 文件检查配置
cat aigc-claw/backend/.env | grep -E "API_KEY|KEY"

# 必需的配置（根据选择的模型）
# LLM: DASHSCOPE_API_KEY / DEEPSEEK_API_KEY / OPENAI_API_KEY / GEMINI_API_KEY
# 图片: ARK_API_KEY / DASHSCOPE_API_KEY
# 视频: DASHSCOPE_API_KEY / VOLC_ACCESS_KEY / KLING_ACCESS_KEY

If the API Key is not configured, you need to remind the user:

  1. Inform them which platform's API Key is missing
  2. Provide how to obtain it
  3. Configuration location (aigc-claw/backend/.env file)
  4. Wait for the user to complete configuration before continuing
Platform API Key variable Obtaining link
DeepSeek DEEPSEEK_API_KEY https://platform.deepseek.com/api_keys
Alibaba Cloud DashScope DASHSCOPE_API_KEY https://bailian.console.aliyun.com/cn-beijing/?tab=home#/home
ByteDance Volcano Ark ARK_API_KEY or VOLC_ACCESS_KEY/VOLC_SECRET_KEY https://www.volcengine.com/product/ark
Kuaishou Kling KLING_ACCESS_KEY/KLING_SECRET_KEY https://klingai.com/cn/dev

🚨 Stop point handling (mandatory rules)

When the query status is stage_completed or waiting_in_stage , you must perform the following steps:

Step 1: Obtain the artifact

curl "http://localhost:8000/api/project/{session_id}/artifact/{stage}"

Step 2: Show it to the user

Display the content in the artifact (option list, suggestions, artifact summary)in fullto the user

Step 3: Ask for a decision

Clearly tell the user:

  • What the options are
  • The meaning of each option
  • What the user needs to choose

Step 4: Wait for the user's reply

Prohibitedto call intervene or continue！

Step 5: Execute after user confirmation

According to the user's choice, call the corresponding API

❌ Incorrect example (the mistake I just made)

收到 suggest_expand 停点 → 直接调用 intervene → 跳过用户确认

✅ Correct example

1. 阶段内部停点触发（如 suggest_expand）
收到 suggest_expand 停点 
→ 获取 artifact 查看内容
→ 展示给用户："系统建议启用创意扩写模式..."
→ 询问："是否同意？"
→ 用户回复"同意" → 调用 intervene

2. 阶段完成停点触发
收到 stage_completed 停点
→ 获取 artifact 查看产物内容
→ 展示给用户："第一阶段已完成，生成了剧本内容..."
→ 询问："是否继续下一阶段？"
→ 用户回复"继续" → 调用 continue

At each stop point, you must：

  1. show the artifact or options to the user
  2. ask for confirmation
  3. continue only after the user confirms

Status determination

status Meaning Operation
idle New session Start project
running Executing Poll and wait
waiting_in_stage Waiting for user intervention Call intervene
stage_completed Stage complete Call continue
session_completed All complete End

Note: Intervention is only needed when status changes; do not repeatedly call the artifact API to "confirm"!

Message Sending Channels

Depending on the channel used to send messages to the user (Feishu/WeChat), read the corresponding reference document under references/send_message/ to get precautions and sending methods:

Task Brief Format

After each phase is completed, the brief sent must include:

  1. What was just completed
  2. What to do next
  3. Content requiring user decision
  4. Web interface link：http://[本地IP]:3000/?session={session_id}&stage={stage}(Note: use the local IPv4 address here, do not use localhost!)
  5. Artifact images/videos (send the files directly; sending only paths is prohibited)

Web Interface Link Format

# 获取本地 IPv4 地址
import socket
local_ip = socket.gethostbyname(socket.gethostname())

# 构造前端 URL
frontend_url = f"http://{local_ip}:3000/?session={session_id}&stage={stage}"

# 发送给用户
send_to_user(f"📊 查看详情：{frontend_url}")

Important: You must use the local IPv4 address (e.g. 192.168.1.x), do not use localhost or 127.0.0.1, otherwise users cannot access it from other devices!

Detailed Reference

According to the user's needs and the current phase, refer to the specific documents under the references/ directory to perform the corresponding operations:

references directory

File Purpose When to View
init_project/ Project Initialization When the user downloads for the first time or requests "initialize project"
init_all.md Complete Initialization Process When the user requests initialization and deployment
init_backend.md Backend Initialization When configuring the backend environment for the first time
init_frontend.md Frontend Initialization When configuring the frontend environment for the first time
run_project/ Project Startup
start_backend.md Starting the backend service When the service is not running
start_frontend.md Starting the frontend service When the service is not running
workflow/ Six-Phase Workflow
create_project.md Create New Project API When starting a new video project
create_script.md Script Generation API When executing the first phase
create_character.md Character/Scene Design API When executing the second phase
create_storyboard.md Storyboard Design API/Plot Continuation API When executing the third phase/when the user requests plot continuation
create_reference.md Reference Image Generation API When executing the fourth phase
create_video.md Video Generation API When executing the fifth phase
create_post.md Post-Production Editing API When executing the sixth phase
modify_character.md Modify Character Prompts When the user requests to modify a character
modify_storyboard.md Modify/Continue Storyboard When the user requests to modify/continue the storyboard
modify_reference.md Modify Reference Image Prompts When the user requests to modify a reference image
modify_video.md Modify Video Prompts When the user requests to modify a video
sandbox/ Temporary Workbench
generate_image_t2i.md Text-to-Image API When the user requests image generation
generate_image_it2i.md Image-to-Image/Style Transfer API When the user requests image style transfer
generate_video.md Short Video Generation API When the user requests to generate a video within 15 seconds
send_message/ Message sending
feishu.md Feishu sending media files When a user initiates a conversation through the Feishu channel and an image/video needs to be sent to the user
wechat.md WeChat sending media files When a user initiates a conversation through the WeChat channel and an image/video needs to be sent to the user