LPM 1.0

lpm-ai.org

Full-duplex character video generator

Generate real-time character videos that speak, listen, sing, and react with 0.35s latency.

FreemiumWebVideo

Visit website

Overview

LPM 1.0 (Large Performance Model) is a 17B-parameter Diffusion Transformer that generates real-time AI character performance video with full-duplex conversation. From a single image input, it produces speaking, listening, singing, and emoting characters with 0.35-second latency and identity consistency across long interactions. It is designed for conversational AI agents, game NPCs, live streaming characters, and accessibility applications.

How it works

LPM 1.0 uses a two-stage architecture. Base LPM is a 17B-parameter Diffusion Transformer trained for controllable, identity-consistent performance through multimodal conditioning. Online LPM is distilled using DMD into a causal streaming generator that compresses diffusion into 2 generation steps, achieving 480P/720P AI video at 24fps with 0.35-second latency.

Key capabilities

LPM 1.0 unifies text, audio, and image control in a single generation pass. Text prompts direct motion and behavior, audio drives lip sync and emotion, and reference images define character identity. It supports zero-shot generalization across any character style, including photorealistic humans, 2D anime, 3D game characters, and non-humanoid creatures, without fine-tuning.

Who it is for

LPM 1.0 serves as a visual engine for conversational AI agents, game NPCs, live streaming characters, and accessibility applications. It is plug-and-play compatible with A2A models like ChatGPT and Doubao, and can generate expressive virtual tutors and companion characters that respond naturally to user interaction in real time.

Pricing and limits

LPM 1.0 pricing starts at $9.9/month when billed annually. Paid plans include credits for AI video generation, private generations, no-watermark exports, and commercial-use rights. Daily free credits are available with no card required. You can cancel anytime, and unused credits remain available until the end of your billing period.

Key features

Real-time streaming
Generates real-time character video with 0.35s latency, 3x faster than alternatives.
Full-duplex conversation
Supports full-duplex conversation with simultaneous speaking and listening video streams.
Identity preservation
Maintains identity consistency for 10+ minutes, with demonstrated sessions up to 45 minutes without drift.
Zero-shot generalization
Accepts any character style as input, including photorealistic, anime, 3D, and non-humanoid, without fine-tuning.
Multimodal control
Unifies text, audio, and image control in a single generation pass for fine-grained directorial control.
Output quality
Generates 480P/720P video at 24fps, supporting both real-time streaming and high-fidelity recording.

Use cases

Conversational AI agents
Transform text-based chatbots into visual conversational agents with human-like character performance.
Game NPCs
Create NPCs that deliver real-time, identity-consistent performance with expressive dialogue and emotional reactions.
Virtual streaming characters
Enable live streaming characters that perform in real time with full-duplex conversation, singing, and emotional expression.
Accessibility and education
Generate expressive virtual tutors and companion characters that respond naturally to user interaction.

Pricing & limits

From the official site · checked 10/10/2026

Starter$9.9/month (billed annually) or $19.9/month (monthly billing)

Advanced$44.9/month (billed annually) or $89.9/month (monthly billing)

Annual Plan$14.9/month (billed annually) or $179/year

See current pricing on the official site

Frequently asked questions about LPM 1.0

What is LPM 1.0 and what does LPM stand for?: LPM 1.0 stands for Large Performance Model. It is a 17B-parameter Diffusion Transformer that generates real-time, full-duplex conversational AI video from a single image input, supporting speaking, listening, singing, and emotional expression at 0.35 seconds latency.
How does LPM 1.0 generate real-time AI video?: LPM 1.0 uses a two-stage architecture. Base LPM is a 17B-parameter Diffusion Transformer trained for controllable, identity-consistent performance through multimodal conditioning. Online LPM is distilled using DMD into a causal streaming generator that compresses diffusion into 2 generation steps, achieving 480P/720P AI video at 24fps with 0.35-second latency.
What character styles does LPM 1.0 support?: LPM 1.0 supports zero-shot generalization across any character style including photorealistic humans, 2D anime, 3D game characters, and non-humanoid creatures. No fine-tuning or domain-specific training is required.
How long can LPM 1.0 maintain identity consistency?: LPM 1.0 maintains identity-consistent AI video generation over unbounded lengths. Demonstrated sessions include 22-minute and 45-minute continuous full-duplex conversations with no identity drift.
Is LPM 1.0 free to try?: Yes. Daily free credits are available with no card required. You can test prompts and preview the workflow before paying.