Ultra-low latency
The fastest response time of any major avatar provider, at 471ms, about the blink of an eye. Read the technical report.
Avatar API for real-time, interactive video
LemonSlice animates any character from a single photo into a real-time avatar that listens, talks, and responds. Add the Avatar API to the stack you already use, or drop a hosted widget into your site with two lines of code.
- Ultra-low latency
- Unlimited custom avatars
- 1000+ concurrency
- 24-hr long calls
[LemonSlice] is state of the art. It's so lifelike. It has been such a pleasure for us at Microsoft to work with LemonSlice.Brad Smith · President of Microsoft
From a single image to a live conversation
An avatar API lets your code create and animate a digital character. The LemonSlice Avatar API goes further and animates that character into a live, two-way conversation. Upload one photo, and go from image to a real-time video call with that character. No training, no fine-tuning, no video shoot.
Every avatar runs on a Character World Model, trained from scratch to listen, talk, and act in real time. It is a step past the deepfake and older generative technology behind traditional real-time avatars.
avatar = lemonslice.AvatarSession(
agent_image_url="https://example.com/image.png",
agent_prompt="Talk with hand gestures."
)What you can build
LemonSlice 2.1 is the default model, and every model runs thousands of concurrent calls. Flash, the Cartoon model, and Pro are there for lower latency, more expressive cartoons, and large-format displays.
Interactive characters for games and entertainment
The Cartoon model is the more expressive option for cartoon companions and NPCs. Real-time image updates change outfits or change scenes without starting a new session.
Tutors for learning and training
Photoreal tutors run on LemonSlice 2.1. The Cartoon model is the more expressive option for cartoon characters. The Action Engine fires gestures and emotion during the call.
Guides for museums, tourism, and airports
Large-format kiosks and exhibits use LemonSlice Pro, the high-resolution model for physical displays, with 24-hour calls.
A face for support, sales, and training
Thousands of concurrent calls, and it is easy to integrate with the voice agent you already use. Flash is the lower-latency model.
Any character. Any style.
LemonSlice is the only interactive avatar model that can animate non-humans, like a turtle or a smiling teacup, as easily as humans.
Update a character's appearance, emotions, and actions during a live session, and swap the source image in under a second.
Built for real-time
Speed and stability are the difference between a demo and production, so the Avatar API is tuned for live conversation.
24-hour live calls
Sessions stay live for up to 24 hours without dropping, built for kiosks, installations, and always-on agents.
Live at 24 fps
Every frame is generated in real time from one photo, so replies feel like a real conversation, not a clip.
Thousands of concurrent calls
Run thousands of simultaneous avatar sessions — built for high-volume production workloads.
How interactive avatars work
LemonSlice adds a face layer on top of your voice agent. It works with any TTS and any LLM, so nothing in your stack has to change. Behind the scenes, speech-to-text and voice-activity detection listen, your LLM replies, text-to-speech generates audio, and LemonSlice turns that audio into real-time character video in a WebRTC call.
Step 1: Pick or upload any image.
Upload a photo of any character. No training, fine-tuning, or character rigs required. If it has a face, LemonSlice can animate it.
ElevenLabs v3
Sonnet 4
OpenAI Voice
Gemini Pro
Cartesia Sonic-3
GPT-5.5 Mini
Gemini Audio
Qwen3 235B
Hume
Step 2: Connect any LLM/voice model.
Plug in any language model, voice provider, and knowledge base through LemonSlice's API. OpenAI, Claude, ElevenLabs, Cartesia, and more.
Step 3: Go live with ultra-low latency.
Deploy to your site with a widget, hosted pipeline, or self-managed pipeline. LemonSlice runs at 471ms p99 latency, so conversations feel instant.
Set up an avatar in 5 minutes with our developer-friendly API.
Add LemonSlice avatars to the custom stack you already use or drop a hosted widget into your website with two lines of code. LemonSlice is built to be API-first.
Choose a code example
async def my_agent(ctx: agents.JobContext):
avatar = lemonslice.AvatarSession(
agent_image_url="https://lemonslice.com/image.png",
agent_prompt="Talk with hand gestures.",
)
session = AgentSession(
stt="deepgram/nova-2",
llm="openai/gpt-5.3-chat-latest",
tts="cartesia/sonic-3",
)
await avatar.start(session, room=ctx.room)
await session.start(
)Let your coding agent add an avatar
Install the LemonSlice skill, then ask Cursor, Claude, or Copilot to add a video avatar to a voice app. The skill checks what you already have. It attaches the avatar to an existing LiveKit agent, or scaffolds the starter app with keys, the voice cascade, an image, and a ringing UI.
$ npx skills add lemonsliceai/skills
Ask your coding agent:
“Add a LemonSlice video avatar to a real-time voice app.”
Models available for every use case
LemonSlice 2.1 is our state-of-the-art model. Available on all plans.
yourwebsite.com
CWM-1 Lite is our low-cost model for consumer interactive avatar apps
- Low cost
- Ultra-low latency
- High volumes
- Instant character creation
LemonSlice Pro is our high-res model for large-format displays and physical installations
- High resolution
- Physical installations
- Exceptional visual quality
- High-fidelity characters
Built for enterprise scale
Multi-region servers
Start the session in the region closest to the client so the media path stays short.
Zero data retention mode
Zero-data-retention mode keeps streaming media and transcripts off disk after a hosted session. Call metadata is still retained.
Pricing
Plans start at $8/mo, and every plan includes API access. The base model runs about $0.16 per included minute. Bring your own LLM or voice model on any tier.
Learn more
