︎

⌘+K      Résumé     Now      Playbook      → 9 years of Everyday 3D art



Go to: Overview Pipeline Workflow Tools Takeaways
Go directly to: Overview Pipeline Workflow Tools Takeaways

Art Critic AI Agent


Building an autonomous, multimodal AI pipeline for daily creative critique and content generation.


Link to Git Repo

Overview

I built an automated AI agent that critiques my daily render artworks in my Everyday Project and outputs shareable video content for social platforms.


Pipeline Breakdown

Stage 1 – Text & Audio Generation


  • Input: Daily render image from my ongoing Everyday project.
  • Perception: The agent analyzes the image using OpenAI’s GPT-4o Vision API.
    • Later replaced by local LLMs (Mistral, Llama3 via Ollama) for a free, offline workflow.
  • Reasoning: Generates an art critique text based on the image content.
  • Voice: Converts the critique text to audio narration.
    • Initially used OpenAI TTS, later integrated free alternatives like Coqui TTS and Piper TTS.
  • Bonus: Suggest an existing artwork related to the artwork critiqued.

Stage 2 – Video Synthesis


  • Combines:
    • Daily render image (as static background)
    • Audio narration
  • Uses FFmpeg (open-source) to produce the final .mp4 video.

Prompt


“Describe and critique this artwork in detail.  Also suggest an existing piece of art that is similar to this based on your analysis. Check and make sure that it is an existing artwork”



Workflow Architecture







Tools & Technologies

  • Text Generation: OpenAI GPT-4o, Mistral/Llama3 via Ollama
  • Text-to-Speech: OpenAI TTS, Coqui TTS, Piper TTS
  • Video Composition: FFmpeg
  • APIs: Meta Graph API, YouTube Data API (explored for auto-posting)


Key Takeaways

  • Created a multi-modal AI Agent exhibiting perception → reasoning → action.
  • Transition from paid APIs to a free, local-first setup, increasing accessibility and sustainability.
  • Open to future extensions:
    • Auto-posting to platforms
    • Engagement-driven feedback loops
    • Fully autonomous daily outputs
Accessibility Statement

I'm committed to making this website fully conformant with WCAG 2.1 level AA. This site relies on HTML, WAI-ARIA, and CSS to work with combinations of assistive technologies and web browsers. If you wish to report an access barrier or have any access requirements, please contact me and I will address it promptly: Go to email.