Talking Head Videos from Topics

I want to build an AI agent that generates avatar talking head videos. User will enter two inputs /variables to begin - a rough description of what the video should be about, and second - the type of avatar image I want (eg woman wearing christmas hat in department store). Then the agent will make the script, voice over, image and then video. Here's how I am thinking the subagents will work - first it will write a script for me (approx. 100-300 words). Secondly, it will turn that script into a voice over audio. Next, it will generate an image of my AI avatar (which should be topical and related to the theme of my video). Finally it will take this custom ai avatar image and the voice over file and generate the talking head video in question, and return the mp4 as the final output. In this way I can create custom AI avatar talking head videos on any topic, and with the avatar image themed to the topic.

First, pre-entered info...

final-output2

The final output will be a single MP4 video file in which a custom AI-generated avatar (based on the user's description) appears to speak the script that was generated and converted into a voiceover.

subagents-extracted2

OVERVIEW OF SUBAGENTS: Generate a talking head MP4 video in which an AI-generated avatar speaks a script created from a user's prompt. SUBAGENT 1: ScriptScribe • Purpose: Generate a 100–300 word script based on the user's prompt. • Skill Used: #171 - Write Voice Over Script Based On Instructions • INPUT: Prompt or topic from user - [variable1] • OUTPUT: Text script (100–300 words) • SUBAGENT FINAL OUTPUT: [script-text] SUBAGENT 2: VocalBuilder • Purpose: Convert the script into an MP3 voiceover. • Skill Used: #170 - Turn Script Into Voice Over MP3 • INPUT: Text script (from Subagent 1) - [script-text] • OUTPUT: MP3 file URL • SUBAGENT FINAL OUTPUT: [voiceover-mp3] SUBAGENT 3: AvatarVision • Purpose: Create an AI avatar image matching the user's description (e.g., "woman wearing a Christmas hat in a department store"). IMPORTANT: Preprompt must specify that avatar should be in 16:9 aspect ratio • Skill Used: #222 - Make Image (Especially With Text) • INPUT: Take user's visual/avatar description - [variable1] and generate an avatar image that matches this. Also, the preprompt for this subagent should explain we want 16:9 aspect ratio (this preprompt will be fixed for this subagent) • OUTPUT: PNG image URL • SUBAGENT FINAL OUTPUT: [avatar-image] SUBAGENT 4: VideoSync • Purpose: Combine the avatar image and the MP3 voiceover into a single lip-synced MP4 video. • Skill Used: #168 - Generate Talking Head Video From MP3 & transcription • INPUT: MP3 URL (from Subagent 2) - [voiceover-mp3] and avatar image (from Subagent 3) - [avatar-image]) • OUTPUT: MP4 video URL • SUBAGENT FINAL OUTPUT: [talking-head-video] SEQUENCE OF ACTIONS: 1. Subagent 1 generates the script (#171). 2. Subagent 2 converts the script into an MP3 (#170). 3. Subagent 3 creates a matching AI avatar image (#222). 4. Subagent 4 produces the final talking head MP4 video (#168).

subagentnumber

[subagentnumber]

Then, agent response...

subagent1refined-new-json-core

G) TASKFLOW SUMMARY Task 1: - ID: #185 - Title task: Generate Initial Script Outline - Description: Creates a structured outline of key talking points based on the user's input prompt, serving as a foundation for the final voice-over script - Input(s) involved: [variable1] - Output token name: [outline-text] Task 2: - ID: #223 - Title task: Convert Outline to Voice-Over Script - Description: Transforms the outline and original prompt into a polished, conversational voice-over script between 100-300 words, optimized for natural speech patterns and engagement - Input(s) involved: [variable1], [outline-text] - Output token name: [script-text] Note: The taskflow shows a sequential process where Task 1 (#185) creates the foundational outline, which is then fed along with the original prompt into Task 2 (#223) to generate the final voice-over script. The [script-text] output from Task 2 becomes the subagent's final deliverable. These tasks align with the two silos mentioned in section F: - Silo 1 corresponds to Task 1 (Outline Creation) - Silo 2 corresponds to Task 2 (Script Generation) Each task has clear inputs and outputs, creating a logical chain from the initial user prompt [variable1] to the final [script-text] output.

subagent1mermaid

<!DOCTYPE html> <html> <head> <title>Mermaid Diagram Example</title> <script src="https://cdnjs.cloudflare.com/ajax/libs/mermaid/9.4.3/mermaid.min.js"></script> </head> <body> <div class="mermaid"> graph TD Start[Start: User Input] --> Input[variable1: Topic/Prompt] Input --> Task1[Task 1: Generate Initial Script Outline] Task1 --> OutlineOutput[outline-text] OutlineOutput --> Task2[Task 2: Convert Outline to Voice-Over Script] Input --> Task2 Task2 --> ScriptOutput[script-text] ScriptOutput --> End[End: Final Script Output] subgraph Silo 1: Outline Creation Task1 OutlineOutput end subgraph Silo 2: Script Generation Task2 ScriptOutput end classDef process fill:#e1f5fe,stroke:#01579b classDef input fill:#e8f5e9,stroke:#2e7d32 classDef output fill:#fce4ec,stroke:#c2185b class Start,Input input class Task1,Task2 process class OutlineOutput,ScriptOutput output class End output </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>

Mermaid Image - https://static.aiz.ac/1738088258-mermaid/mermaid1.png

prompts-required

[prompts-required]

subagent1-json-array

[ { "taskID": 185, "semanticTitleOfTask": "Generate Initial Script Outline", "taskDescription": "Generates a concise, structured outline of key talking points from the user’s prompt, serving as the foundation for a longer voice-over script. This ensures the final script flows logically and remains focused on the primary topic.", "inputDescription": "Requires the user's main topic or prompt ([variable1]) to formulate a brief outline including 3-4 major points. These points should naturally connect and help shape the final voice-over script.", "inputRequired": ["variable1"], "outputDescription": "Produces a short outline capturing the primary points relevant to the user’s prompt, setting the stage for the next step of script generation.", "outputName": "outline-text", "promptInstruction": "You are a professional content strategist creating an outline for a voice-over script. Using the provided topic, create a brief, structured outline that includes 3-4 main talking points. The outline should be concise yet comprehensive enough to serve as a foundation for a 100-300 word voice-over script. Focus on points that will engage viewers and flow naturally in spoken form. Each point should logically connect to the next, creating a cohesive narrative structure." }, { "taskID": 223, "semanticTitleOfTask": "Convert Outline to Voice-Over Script", "taskDescription": "Transforms the structured outline and the user’s original input into a polished, conversational voice-over script of 100-300 words. The resulting script is engaging, easy to follow, and maintains a natural spoken flow.", "inputDescription": "Requires the original user prompt ([variable1]) and the structured outline ([outline-text]) generated by Task 1. Both inputs are combined to ensure the final script accurately reflects the user’s topic while incorporating the logical flow from the outline.", "inputRequired": ["variable1", "outline-text"], "outputDescription": "Produces a final text-only voice-over script, 100-300 words, designed to be naturally spoken aloud and to capture the audience’s attention through clear organization and a conversational style.", "outputName": "script-text", "promptInstruction": "You are a professional voice-over script writer. Using the provided outline and original topic, create a natural, conversational script of 100-300 words. The script should be engaging, flow naturally when spoken aloud, and maintain a friendly, approachable tone. Format the script using proper punctuation and natural breaks. Avoid complex sentences or words that might be difficult to pronounce. The script should hook viewers early, develop key points clearly, and end with a strong conclusion. Use contractions and informal language where appropriate to maintain a conversational feel." } ]

subagent1-json-array-mermaid

<!DOCTYPE html> <html> <head> <title>Mermaid Diagram Example</title> <script src="https://cdnjs.cloudflare.com/ajax/libs/mermaid/9.4.3/mermaid.min.js"></script> </head> <body> <div class="mermaid"> graph TD A[User Input: variable1] --> B[SubAgent: ScriptScribe] B --> C[Task 1: Generate Initial Script Outline] C --> D[Output: outline-text] D --> E[Task 2: Convert Outline to Script] A --> E E --> F[Output: script-text] F --> G[Final Output: 100-300 Word Voice-Over Script] subgraph SubAgent Flow B C D E F end style A fill:#f9f,stroke:#333 style G fill:#9f9,stroke:#333 style B fill:#ddf,stroke:#333 style SubAgent Flow fill:#f5f5f5,stroke:#666 </div> <script> mermaid.initialize({ startOnLoad: true }); </script> </body> </html>