> ## Documentation Index
> Fetch the complete documentation index at: https://lemonslice.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Welcome to LemonSlice
Build interactive avatars that can listen, talk and respond directly inside your product or website using the LemonSlice API.
✓ [Ultra-low latency](https://lemonslice.com/blog/lemonslice-flash)\
✓ 1000+ concurrency \
✓ Multi-hour long calls\
✓ Instant avatar creation (1 photo) \
✓ 5-minute API setup
Build a production-ready avatar app with Next.js + LiveKit in 5 minutes
## **The most advanced interactive avatar model**
[**LemonSlice**](https://lemonslice.com/) is an AI research lab that built the world’s first **Character World Model**: interactive characters that talk, listen, and react in real-time based on your conversations. Character World Models are more advanced than traditional real-time avatars, which are based on deepfake or older generative technology.
Here are some of the capabilities of the LemonSlice model:
Create any character from a **single image**. No training or fine-tuning required. Go from image to live video call with that character instantly.
## How it works
Every real-time conversation with a character goes through the following steps:
1. A speech-to-text (**STT**) and voice-activity detection (**VAD**) model listen to the user
2. An **LLM** decides what to say back
3. Text-to-speech (**TTS**) turns the reply into audio
4. **LemonSlice** turns the audio into a real-time video of the character speaking within a live video call
All of this runs within a harness that manages the orchestration and WebRTC connection. LemonSlice connects to all major WebRTC providers to make orchestration easy.
The inputs to the LemonSlice API are:
1. **Image** - defines the appearance of your character.
2. **Audio** - streaming audio from a TTS model like ElevenLabs or Cartesia. This audio is exactly what your character will say.
3. (optional) **Action Engine** - determines what actions (waving, holding a phone, looking away) the character should take during the conversation. Currently Enterprise only.
The quality of your avatar is *heavily* determined by your image. See [Avatar Image Tips](https://lemonslice.com/docs/prompting-guide/avatar-image-tips) for framing and posing tips.
## **Integrate real-time avatars into your app**
Add LemonSlice to the stack you already use. LemonSlice integrate with all major WebRTC providers like LiveKit, Daily/Pipecat, and Agora. These providers all have Conversational AI toolkits that make it simple to combine any LLM, TTS, and STT component with LemonSlice.
1. **Pick a framework** — follow the [LiveKit integration](/livekit) or [Pipecat integration](/pipecat) guide.
2. **Build your UI** — run your own call lifecycle and frontend around the avatar session. See [Production checklist](/reference/production-checklist) for call lifecycle patterns.
You can think of LemonSlice like a graphics layer or “face layer” that’s added on top of your voice agent. It is compatible with any TTS and LLM model.
Or, use one of the low-code options below to get started with minimal engineering effort.
The [Widget](/widget/overview) is a no-code solution for adding an avatar to your website. It's like embedding a YouTube video - just 2 lines of code. LemonSlice takes care of everything else.
The [Hosted Pipeline](/hosted/overview) is a low-code solution. LemonSlice manages the entire avatar and agent infrastructure. You build the UI for the avatar call.
Here's a comparison of the options:
| | **Self-managed pipeline** | **Hosted pipeline** | **Widget** |
| ----------------------- | ----------------------------- | ---------------------------- | ------------------------------------- |
| **Complexity** | ⚙️⚙️⚙️ Full code | ⚙️⚙️ Low code | ⚙️ No code |
| **You control** | Speech, Intelligence, Call UI | Call UI | Nothing |
| **LemonSlice controls** | Avatar | Speech, Intelligence, Avatar | Speech, Intelligence, Avatar, Call UI |
## Try the live demo
Full demo on lemonslice.com
The inputs to the LemonSlice API are:
1. **Image** - defines the appearance of your character.
2. **Audio** - streaming audio from a TTS model like ElevenLabs or Cartesia. This audio is exactly what your character will say.
3. (optional) **Action Engine** - determines what actions (waving, holding a phone, looking away) the character should take during the conversation. Currently Enterprise only.