Craft Generative AI for User Generated Content
Inspired by vision to include generative AI-powered user-generated content (UGC), this challenge tasks participants with building a sophisticated multi-agent system for creating, moderating, and personalizing short-form multimodal content. The system will employ Claude Opus 4.1 for its advanced creative generation and nuanced ethical reasoning capabilities, ensuring content aligns with user defined brand safety and creative guidelines. LangGraph will be used to orchestrate a dynamic, graph-based workflow for content generation, review, and iteration. This will include agents specializing in ideation, multimodal asset generation (text-to-video/image), and a crucial moderation agent that applies 'Guidance' for constrained generation, ensuring content adheres to specific brand safety and thematic rules. The challenge emphasizes balancing creative freedom with strict content policy enforcement.
What you are building
Develop a LangGraph-orchestrated multi-agent system using Claude Opus 4.1 and Guidance for generating and moderating multimodal UGC.
How work is evaluated
The evaluation will focus on the system's ability to generate high-quality, Disney-appropriate multimodal content while strictly adhering to moderation guidelines. It will assess the LangGraph workflow's efficiency and the effectiveness of Guidance in enforcing constraints.
Datasets
Public datasets, plus the private datasets you have access to.
Learning goals
Implement multimodal content generation (text, image, short video) using Claude Opus 4.1 and Llama 4.
Utilize Guidance to enforce brand safety and thematic constraints during content generation and review.
Orchestrate a multi-agent system for iterative content refinement and automated moderation against Disney+ guidelines.
Design a LangGraph workflow for generative AI content creation and moderation.
Resources
Sample data for 2 tasks
Challenge setup
The evaluation will focus on the system's ability to generate high-quality, brand-appropriate multimodal content while strictly adhering to moderation guidelines. It will assess the LangGraph workflow's efficiency and...
Challenge input
{ 'user_prompt': 'A cheerful short story about Mickey Mouse discovering a hidden magical garden.' }
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ 'generated_text': '...', 'generated_image_url': '...', 'video_concept': '...', 'moderation_status': 'approved|rejected', 'moderation_report': '.....
- The LangGraph workflow completes successfully, generating and moderating content without errors.
- The moderation agent correctly flags at least 80% of intentionally violating content using Guidance.
- Generated content adheres to general family-friendly and positive themes.
- Creative Diversity target: 0.7
- Python execution harness
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- The LangGraph workflow completes successfully, generating and moderating content without errors.
- The moderation agent correctly flags at least 80% of intentionally violating content using Guidance.
- Generated content adheres to general family-friendly and positive themes.
Metrics
- Creative Diversity · Target 0.7
- Constraint Efficacy · Target 0.85
Workflow tasks
- Task 1
Input: { 'user_prompt': 'A cheerful short story about Mickey Mouse discovering a hidden magical garden.' }
Output: { 'generated_text': '...', 'generated_image_url': '...', 'video_concept': '...', 'moderation_status': 'approved|rejected', 'moderation_report': '...' }
Execution harness: Python
- Task 2
Input: { 'content_to_moderate': {'text': '...', 'image_concept': '...', 'video_concept': '...'}, 'rules': ['no violence', 'no copyrighted characters not from IP owners'] }
Output: { 'violations_detected': ['rule_id_1'], 'justification': '...' }
Execution harness: JavaScript
Recipe freeze marker
No recipe freeze time is recorded.