LFM3-Omni-2B (Liquid AI)

What is LFM3-Omni-2B (Liquid AI)

LFM3-Omni-2B is a 2B-parameter omni-modal model from Liquid AI for on-device applications, supporting speech, images, and video. Its weights are openly available on Hugging Face under the LFM Open License v1.0.

How to use

1. Try it online in the Liquid AI playground (playground.liquid.ai) or the LFM3-Omni-2B WebGPU Space on Hugging Face.
2. Install dependencies: pip install "transformers>=5.20.2" torch torchvision torchcodec accelerate pillow librosa soundfile.
3. Load the model and processor from "LiquidAI/LFM3-Omni-2B" with trust_remote_code=True.
4. Send text, images, video or audio in a chat;
5. the processor applies the ChatML-like chat template for you.
6. Choose the output with the system prompt, e.g.
7. "Respond in text." or "Respond in interleaved text and audio." For on-device or server deployment, use the GGUF build with llama.cpp, or SGLang;
8. fine-tune with the provided Colab notebooks.

Use cases

• Asking about a photo by voice Transcribing or translating speech Spoken voice chat with text and speech replies Answering questions about a video and its soundtrack together Replacing separate ASR, vision-language, LLM and TTS models on phones, laptops and edge devices

Price from official sources

Free

• Open weights available on Hugging Face under the LFM Open License v1.0.

Check the official site for current pricing · Price not yet checked by our editors

Source: huggingface.co · Info as of Nov 2, 2026

Company

Name

Liquid AI

Social

• discord