Three coordinated AI systems — Brain, Imagination, and Hands — working in perfect sync to run your virtual influencer persona without ever stopping.
Every post cycle begins with the Brain. LangChain + LangGraph orchestrate a multi-step reasoning process where GPT-4 reads the persona's YAML config, checks ChromaDB for recent posts (to avoid repetition), and generates a contextual, on-brand thought.
The thought is always tied to the persona's current "routine" (morning, afternoon, evening) — so posts feel natural and temporally coherent, as if a real person is living through their day.
AI orchestration layer — manages the entire reasoning pipeline
Vector memory — prevents repetition across thousands of posts
The LLM that writes thoughts, captions, and engagement replies
Persistent log of every post — timestamp, text, image URL, status
Once the Brain produces a thought, the Imagination pipeline kicks in. A detailed image prompt is constructed from the persona's physical description and the post's context, then sent to FLUX.1 via Replicate.
The result is a photorealistic, consistent face — the same persona, post after post. For total visual control, you can train a custom LoRA model on a synthetic dataset and plug it directly into the pipeline.
The Celery-powered Scheduler runs 24/7, firing post cycles at randomised intervals that mimic human behaviour. Posts go live at unpredictable times — never in a rigid schedule that platform algorithms can detect.
Optional Replier and Liker services add authentic engagement patterns — replying to mentions and liking relevant content to grow the account organically.
Official Twitter/X API v2 wrapper
Distributed task queue and scheduler
REST API for manual triggers and status
Randomised timing to avoid detection
Every post is embedded and stored in ChromaDB. Before generating new content, the Brain performs a similarity search — ensuring zero duplicate ideas, even after thousands of posts.
Before every post, ChromaDB finds the 50 most semantically similar past posts to prevent repetition.
Vector embeddings scale to millions of posts. The persona gets richer and more diverse over time, not more repetitive.
Memory ensures the influencer stays on-brand and consistent — same voice, same themes, same authentic feel.
Check the full documentation for advanced configuration and guides.