LemonSlice

LemonSlice: Live video avatars you can talk to, built from one photo

LemonSlice turns a single image into a real-time, interactive AI video avatar. Upload a corporate headshot, a cartoon mascot, or even a painting, and the LemonSlice-2 model starts a live video conversation immediately — no training footage, no clone setup, no style restrictions. The company pitches the product as a face layer for voice agents: you bring the speech-to-text, LLM, and voice stack, and LemonSlice renders a lip-synced, gesturing character on top of it.

Under the hood, LemonSlice-2 is a 20-billion-parameter causal diffusion model that generates roughly 20 frames per second on a single GPU and streams video over WebRTC, with infinite-length generation that avoids error accumulation on long calls. Developers integrate through a REST API, an embeddable website widget, or self-managed pipelines built on LiveKit, Pipecat, or Daily. Plans start at $7 per month billed annually ($8 billed monthly), with usage metered in credits per interactive call minute and unused credits rolling over.

Key Features

• Zero-shot avatars from one photo — photoreal faces, cartoons, mascots, and stylized art all work without fine-tuning
• Real-time generation: a 20B-parameter causal diffusion model streaming about 20 fps per GPU over WebRTC
• Every pixel generated live, including facial emotion, hand gestures, body movement, and mid-call appearance changes
• Developer API with session management, motion and emotion triggers, and function calling
• Embeddable widget that adds a talking avatar to any site with one line of code, with guides for Shopify, Squarespace, Wix, and Next.js
• Works with your own STT/LLM/TTS stack via LiveKit and Pipecat; ElevenLabs-backed voices with 5–60 custom voices per plan

Ideal Use Case

LemonSlice fits teams that already run voice agents — customer support, tutoring and language learning, e-commerce shopping hosts, corporate training — and want to give them a visible, expressive face without building a real-time video pipeline from scratch. The widget path also suits marketers who want a conversational avatar on a site with minimal engineering.

How LemonSlice differentiates

LemonSlice (Y Combinator W24) announced a $10.5M seed round in December 2025 from Matrix Partners, Y Combinator, Dropbox co-founder Arash Ferdowsi, former Twitch CEO Emmett Shear, and The Chainsmokers. Unlike template- or warping-based avatar tools, it generates every pixel from its own diffusion model, which is why it can animate non-human and cartoon characters, and it publishes per-minute pricing openly. A life-size conversational exhibit at the Theodore Roosevelt Presidential Library runs on the platform.

FAQ

Do I need training video to create an avatar? No. LemonSlice is zero-shot: a single image is enough to start a live conversation. Is there a developer API? Yes — a REST API plus LiveKit and Pipecat integrations, and an embeddable web widget for no-code deployment. How much does it cost? Paid plans start at $7/month billed annually ($8 monthly), with per-minute credit pricing and credit rollover. Can it animate cartoons or paintings? Yes. The model handles photoreal headshots, cartoons, mascots, and artwork equally.

tl;dr

LemonSlice turns one photo into a live, talking AI video avatar, delivered as an API, widget, or hosted pipeline — a $10.5M-backed alternative to HeyGen, Tavus, and D-ID for interactive avatars.

Related

Looking for more options? Browse the Video Creation directory or read our best AI video tools listicle. LemonSlice is also tracked on Crunchbase.