Editorial illustration for LPM 1.0 creates 45‑minute lip‑synced video from a single photo in real time
LPM 1.0: AI Turns Single Photo into 45-Min Video
LPM 1.0 creates 45‑minute lip‑synced video from a single photo in real time
Every few months, a new AI demo promises easy fake video. LPM 1.0 might actually deliver. Give it one photo and an audio track.
It spits out a lip-synced video instantly, running stable for 45 minutes. The source can be a real person, an anime drawing, or a video game character. No special training required.
It works by streaming live instead of rendering a whole file. Its clever trick? Alongside your main photo, you feed it extra reference shots—different angles, different expressions.
So it never invents your teeth or guesses how your wrinkles form. It just pulls those details from the library you provided. The model listens to audio to create reactive nods and eye movements.
It drives lips and body to match speech. In silent moments, simple text prompts generate idle behavior. This isn't an incremental update.
It's a different way to build a talking head.
LPM 1.0 works across different image styles, photorealistic faces, anime, and 3D game characters, without any additional training. The entire video generation runs as a streaming process in real time rather than rendering a finished video all at once. Videos up to 45 minutes long should remain stable.
LPM 1.0 utilizes what the researchers call "multi-granularity identity conditioning:" alongside a main image, the model also receives reference images from different angles and with varying facial expressions. This means it doesn't have to invent details like teeth, wrinkles tied to specific emotions, or profile views on its own -- it can pull them directly from the reference material. When listening, it generates reactive facial expressions like nodding or gaze shifts based on incoming audio.
When speaking, the response audio drives lip movements and body language. During pauses, LPM generates natural idle behavior based on text instructions. Beyond real-time conversation, LPM 1.0 also supports offline video generation from existing audio, useful for podcasts or movie dialogs, according to project manager Ailing Zeng.
The core advance is straightforward: the model isn't generating a face from scratch. It's assembling one from known parts. That sidesteps the weird, melting look plaguing other systems when they imagine unseen details.
The 45-minute runtime is a technical flex, proving it won't glitch out or desync during a long conversation. That stamina, combined with real-time streaming, points directly to live applications—customer service avatars, virtual tutors, streaming overlays for gamers. The offline mode for dubbing podcasts?
Almost an afterthought, but a practical one. The barrier for creating a persistent, convincing digital person just dropped. We'll see how far it falls.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives Inside the .claude Folder: How AI Stores Its Working State Unannounced
From Archives Analysis of Four Core Dilemmas Highlights AI Agents' Security Risks
Common Questions Answered
How does LPM 1.0 generate video from a single photo across different visual styles?
LPM 1.0 uses a 'multi-granularity identity conditioning' technique that allows it to generate videos from a single image across photorealistic faces, anime, and 3D game characters without additional training. The model can create up to 45-minute videos that maintain the original image's identity while synchronizing lip movements and displaying natural emotional variations.
What makes LPM 1.0's video generation process unique compared to traditional rendering methods?
Unlike traditional video rendering that requires batch processing and extensive computational resources, LPM 1.0 generates videos as a streaming process in real time. This approach allows for continuous video generation with smooth transitions and the ability to create lengthy videos up to 45 minutes long without needing a render farm.
How does LPM 1.0 integrate with voice AI technologies like ChatGPT?
LPM 1.0 can directly hook into voice AI systems like ChatGPT to produce speaking avatars that display nuanced behaviors such as hesitation, gaze shifts, and emotional changes. The model can generate avatars that not only lip-sync with spoken content but also provide a more natural and dynamic conversational experience across various visual styles.
