AI Virtual Influencer Engine · ~$8/mo · Open Source

Your virtual influencer.
Runs itself.

Write a YAML file. Zirelia builds a virtual persona with a consistent photorealistic face, writes original posts, and publishes to Twitter/X around the clock — growing a following while you sleep. Runs on a Raspberry Pi for ~$8/month.

Star on GitHub Quick Start

Python 3.10+ Docker Ready GPT-4 Brain FLUX.1 Images

Sienna Fox AI-Generated

@sienna_fox_ai · just now

morning glow hits different when you stayed up editing until 3am ✨
some days the grind is the glamour

🎨

FLUX.1 Generated

Sienna Fox — California lifestyle

replicate/flux-1.1-pro

2.4k 841 127

Posted by Zirelia

Scheduler running · next post in 4h 22m

GPT-4o-mini · ChromaDB ✓

⚠️

Important — Read Before Starting

Starting the bot on a new account without proper warm-up is the #1 cause of permanent shadowbans.

Read the Anti-Ban Guide →

How it works

What is Zirelia?

A fully autonomous virtual influencer engine. You define the persona once — it runs forever.

🧠

Thinks

GPT-4 generates contextual, on-brand thoughts based on the persona's daily routine and ChromaDB memory of past posts — no repetition.

🎨

Visualises

FLUX.1 via Replicate generates a consistent, photorealistic face for every post. Optional custom LoRA for total visual control.

📤

Posts

A smart Celery scheduler publishes content autonomously to Twitter/X at randomised, human-like times. Looks completely organic.

💬

Engages

Optional services reply to mentions and like relevant content, growing the account authentically while you sleep.

Reference persona: Sienna Fox — real account running on Zirelia since January 2025. Read the case study →

📍 Current Status (February 2026)

Component	Technology	Status
🐦 Social Platform	Twitter / X	✅ Production Ready
🎨 Image Generation	FLUX 1 via Replicate	✅ Production Ready
🧠 LLM Brain	OpenAI GPT-4 / GPT-4o-mini	✅ Production Ready
💾 Memory	ChromaDB (RAG)	✅ Active
🖥️ Control Dashboard	—	🔴 Planned — Phase 3
📸 Instagram / Others	—	🔴 Planned — Phase 4

Architecture

🏗️ How the engine works

Three coordinated systems, one seamless pipeline.

🧠

Brain

The thinking engine. Generates ideas, writes posts, and remembers what was already said.

LangChain + LangGraph

OpenAI GPT-4 / GPT-4o-mini

ChromaDB (vector memory)

🎨

Imagination

The visual pipeline. Creates a consistent, photorealistic face for every single post.

Replicate API

FLUX.1 / SDXL + custom LoRA

Visual QA pipeline

📤

Hands

The publishing system. Posts at human-like times, replies to mentions, likes relevant content.

Tweepy (Twitter/X API)

Celery + Redis scheduler

Anti-detection timing

Full Tech Stack

Python 3.10+

Language

LangChain / LangGraph

AI orchestration

FLUX.1 / Replicate

Image generation

Tweepy

Twitter/X API

FastAPI + Uvicorn

REST API

Celery + Redis

Task queue

PostgreSQL 15

Database

ChromaDB

Vector memory (RAG)

Docker Compose

Deployment

Quick Start

🚀 Up and running in 5 steps

Prerequisites: Docker Desktop, Git, and API keys for OpenAI + Replicate + Twitter/X

Clone the repository

                    git clone https://github.com/zirelia/zirelia.git

                    cd zirelia

Configure your API keys

                    cp .env.template .env

                    # Fill in your API keys:

                    OPENAI_API_KEY=sk-...

                    REPLICATE_API_TOKEN=r8_...

                    REPLICATE_MODEL_VERSION=black-forest-labs/flux-1.1-pro

                    TWITTER_API_KEY=...

                    TWITTER_API_SECRET=...

                    TWITTER_ACCESS_TOKEN=...

                    TWITTER_ACCESS_TOKEN_SECRET=...

                    TWITTER_BEARER_TOKEN=...

Launch with Docker

                    docker compose up --build -d

                    # Starts: api · scheduler · worker · db · redis

Verify it's running

                    docker compose ps

                    docker compose logs scheduler -f

Post manually (optional)

                    # Dry run — no actual posting:

                    python main.py --platform twitter --dry-run

                    # Post now with image:

                    python main.py --platform twitter --mode hybrid

                    # Text only, custom topic:

                    python main.py --platform twitter --mode text --topic "morning coffee ritual"

config/persona.yaml

Sienna Fox

                        name: "Sienna Fox"
                        age: 23
                        nationality: "American (Los
                            Angeles, CA)"

                        physical_traits:
                        hair: "Light Brown / Dirty Blonde
                            (Wavy)"
                        eyes: "Hazel / Warm Brown"
                        vibe: "Natural, Sun-kissed"

                        traits:
                        - "Confident"
                        - "Playful / Flirty"
                        - "Witty"

                        voice:
                        tone: "Seductive but
                            friendly"
                        style: "Short, punchy, uses emojis
                            ✨ 😉"

                        routine:
                        morning:
                        - "Waking up in silk sheets"
                        - "Morning stretch"
                        evening:
                        - "Sunset at the beach"
                        - "Glass of wine"
                    

🎭 Persona

Your persona lives in one YAML file.

No code. No prompt engineering. Just describe who your influencer is — personality, look, voice, daily routine — and Zirelia does the rest. Forever.

Name, age, nationality, physical traits
Personality traits and tone of voice
Daily routine for contextual posts
Custom LoRA model for visual identity
ChromaDB prevents post repetition

Read Config Guide →

Roadmap

🗺️ Where we're going

✅

Phase 1 Complete

Foundation: Twitter, FLUX.1, Smart Scheduler, ChromaDB memory

🟡

Phase 2 In Progress

Infrastructure: PostgreSQL pgvector, Celery integration, engagement automation

🔴

Phase 3 Planned

Control Dashboard (Streamlit/React) — manage persona, review posts, analytics

🔴

Phase 4 Planned

Multi-Platform: Instagram, LinkedIn, Telegram

🔴

Phase 5 Planned

Multimedia: ElevenLabs voice, video clips (Reels/TikTok)

🐳 Docker

All services in one command

docker compose up --build -d

api

FastAPI REST server (port 8000)

scheduler

Autonomous posting daemon — runs 24/7

worker

Celery background task worker

PostgreSQL 15 (persistent storage)

redis

Message broker / task queue

replier

Reply to Twitter mentions (manual profile)

--profile manual

liker

Like relevant content (manual profile)