E-Fantasy

Private AI roleplay companion powered by Venice.ai

NOTE: The demo video shows the censored version so we don't get banned by Youtube/Devpost :) Try it yourself for the uncensored version ;)

E-Fantasy is a fully private, on-device AI companion that combines Venice.ai's uncensored LLM chat, image generation, and video generation into an immersive roleplay experience. Your data never leaves your browser. Your API key never touches a database. Zero telemetry, zero logging, zero judgment.

Why This Matters

Roleplay is the #1 consumer use case for AI that nobody in Silicon Valley wants to talk about. On OpenRouter, character/roleplay models consistently dominate the usage charts — often outpacing coding and productivity tools combined. Character.ai hit 20M+ MAU faster than almost any consumer app in history. The demand is massive, real, and growing.

But every existing platform has the same fatal flaw: your most intimate conversations sit on someone else's server. Character.ai reads your chats. Replika stores everything. Even "private" apps phone home. Users know this, and it creates a chilling effect — people self-censor with their own AI companions because they don't trust the platform.

Venice.ai is the only infrastructure that solves this. True privacy-first AI inference means E-Fantasy can offer something no competitor can: a roleplay companion where the user's data genuinely stays private. Not "we promise we won't look" private. Architecturally private.

How It Works

E-Fantasy is a BYOK (Bring Your Own Key) web app. Users enter their Venice API key once, and it's stored in their browser's localStorage — never sent to any server except Venice's own API. The app's backend is a set of stateless proxy routes that forward requests to Venice and store absolutely nothing.

The Full Venice Stack

E-Fantasy uses three Venice API capabilities in a single coherent experience:

1. LLM Chat (glm-4.7-flash-heretic) — Uncensored, streaming roleplay conversation with rich character personas. The model stays in character, drives scenes forward, and naturally describes visual moments.

2. Image Generation (lustify-v7) — When the AI describes a visual scene, it tags it with [IMAGE: ...]. The app automatically generates that scene as an inline image, creating a visual novel experience.

3. Video Generation (wan-2.6-text-to-video) — For dramatic moments, the AI uses [VIDEO: ...] tags. The app generates a still image first, then animates it into a short video clip — chaining image and video generation seamlessly.

This makes E-Fantasy a showcase for Venice's full multimodal pipeline: LLM describes → image generates → video animates, all in one conversation flow.

Privacy Architecture

[Browser]                          [Venice API]
   │                                    │
   │  API key in localStorage           │
   │  IndexedDB for all data            │
   │                                    │
   ├──── /api/chat ────────────────────►│  Chat completions (streaming)
   ├──── /api/image ───────────────────►│  Image generation
   ├──── /api/video/queue ─────────────►│  Video generation
   └──── /api/video/retrieve ──────────►│  Video retrieval
                                        │
         Stateless proxy routes         │
         Store nothing                  │
         Log nothing                    │

• All user data (personas, conversations, memories, generated media) lives in the browser's IndexedDB
• The API key lives in localStorage
• Route handlers are pure proxies — they forward the key from a request header and return the response
• No database, no auth system, no user accounts — there's nothing to breach
• Full data export/import — users own their data as a JSON file

Features

• Persona System — Create detailed characters with appearance, personality, backstory, scenarios, and preferences. Each persona maintains separate conversation history and memories.

• Persistent Memory — The app periodically extracts key facts from conversations (preferences, events, relationship dynamics) and injects them into future prompts. Characters remember what matters.

• Inline Media Generation — Images and videos appear directly in the chat flow, creating an illustrated story experience rather than a separate generation workflow.

• Gallery View — Browse all generated media for any persona in a grid view.

• Demo Mode — One toggle switches all content to SFW for presentations and demos.

• Data Portability — Export everything as JSON. Import it back. Delete everything. Full user control.

The Market Opportunity for Venice

The AI companion/roleplay market is a multi-billion dollar opportunity that mainstream AI providers actively refuse to serve:

• Character.ai: 20M+ MAU, $1B+ valuation, but increasingly censored and all data on their servers
• Replika: 30M+ downloads, recently forced to remove intimate features under pressure
• OpenRouter roleplay traffic: Consistently the largest single category of API usage
• Chai AI, Crushon, JanitorAI: Millions of users across fragmented, often unreliable platforms

Every one of these users would prefer a private, uncensored experience if it existed. Venice.ai is uniquely positioned to be the infrastructure layer that powers this — and E-Fantasy demonstrates exactly how that works as a polished end-user product.

Every E-Fantasy session is Venice API usage. Chat tokens streaming. Images generating. Videos rendering. The more users engage, the more Venice infrastructure they consume — and these users engage a lot. Roleplay sessions average 10-50x more tokens per session than typical chatbot interactions.

Built With

• venice