Multi-Agent AI for Dynamic Short-Form Video Generation
Design and implement a multi-agent system using AutoGen that simulates a creative studio for generating short-form video content (e.g., YouTube Shorts). Inspired by YouTube's vision for AI-generated media, this system will coordinate specialized agents like a 'Scriptwriter Agent,' 'Visuals Generator Agent,' and 'Editor Agent' to autonomously conceptualize, script, and outline visual elements for a short video based on a user's prompt. The challenge emphasizes complex multi-agent conversations, conditional workflows, and the integration of diverse AI models for different generation tasks, with human-in-the-loop validation at critical stages. This system should be capable of producing a detailed production plan and asset descriptions, even if it doesn't render the final video.
What you are building
The core problem, expected build, and operating context for this challenge.
Design and implement a multi-agent system using AutoGen that simulates a creative studio for generating short-form video content (e.g., YouTube Shorts). Inspired by YouTube's vision for AI-generated media, this system will coordinate specialized agents like a 'Scriptwriter Agent,' 'Visuals Generator Agent,' and 'Editor Agent' to autonomously conceptualize, script, and outline visual elements for a short video based on a user's prompt. The challenge emphasizes complex multi-agent conversations, conditional workflows, and the integration of diverse AI models for different generation tasks, with human-in-the-loop validation at critical stages. This system should be capable of producing a detailed production plan and asset descriptions, even if it doesn't render the final video.
Shared data for this challenge
Review public datasets and any private uploads tied to your build.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
PlanCompleteness
All required components of the production plan (title, logline, script scenes, asset list) are present.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
PromptAdherence
Generated plan aligns with the user's initial prompt and constraints (audience, length, theme).
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
AgentCommunicationFlow
Agents engage in a logical and coherent conversation to reach the content generation goal.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
HumanFeedbackIntegration
Human feedback is correctly processed and results in appropriate, relevant plan revisions.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Creativity_Score
Subjective score for the originality and entertainment value of the generated plan (1-5, 5 being highly creative). • target: 4 • range: 1-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
VisualFeasibility_Score
How realistic and detailed the visual descriptions are for actual video production (1-5, 5 being highly feasible). • target: 4 • range: 1-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Iteration_Efficiency
Number of turns required for agents to effectively address human feedback (fewer is better, target around 2-3 turns). • target: 2 • range: 1-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master AutoGen's ConversableAgent and UserProxyAgent to create complex multi-agent conversations.
Implement advanced agent roles using Claude Opus 4.1 for creative conceptualization and script generation.
Leverage OpenAI o3 for specialized tasks such as generating visual asset descriptions or catchy titles.
Design and integrate custom tools for agents to simulate external media generation services.
Implement dynamic workflow branching in AutoGen based on agent outputs or human feedback.
Explore how DeepSpeed can optimize fine-tuned models for specific content generation sub-tasks within the agent workflow.
Orchestrate output storage of generated plans and assets using Azure Blob Storage.
How this agent runs
Evaluation will assess the generated production plan's completeness, creativity, adherence to the prompt, and feasibility of proposed visual assets.
Challenge input
{"user_prompt": "str", "target_audience": "str", "video_length_seconds": "int"}
Agent execution
The configured agent processes the input under the challenge policy.
Evaluated output
{"title": "str", "logline": "str", "script_scenes": [{"scene_number": "int", "dialogue": "str", "visuals": "str", "sound_cues": "str"}], "asset_lis...
- All required components of the production plan (title, logline, script scenes, asset list) are present.
- Generated plan aligns with the user's initial prompt and constraints (audience, length, theme).
- Agents engage in a logical and coherent conversation to reach the content generation goal.
- Creativity Score target: 4
- 2 public reference cases
- Python execution harness
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- PlanCompleteness · Weight 1
- PromptAdherence · Weight 1
- AgentCommunicationFlow · Weight 1
- HumanFeedbackIntegration · Weight 1
- Creativity_Score · Weight 1
- VisualFeasibility_Score · Weight 1
- Iteration_Efficiency · Weight 1
Recipe state
This is a preview. The configuration can change before the evaluation recipe is locked.
Run this agent on your dataset and AI stack
Bring your dataset, model providers, and success criteria. We will scope the right managed run for your team.
Scope a managed run[ok] Wrote CHALLENGE.md
[ok] Wrote .versalist.json
[ok] Wrote eval/examples.json
Requires VERSALIST_API_KEY. Works with any MCP-aware editor.
DocsFind another challenge
Jump to a random challenge when you want a fresh benchmark or a different problem space.