# LFM3 Omni: A Tiny Local AI That Sees and Hears

A companion to the LFM3 Omni Short: a chart and a recording in one prompt file, and the chart mistake it fixed after feedback.

[HTML](https://lukejbyrne.com/library/guides/00hck2_hbju/)
Content language: en
Author: Luke Byrne
Video published: null
Guide published: 2026-11-04

[Watch the video](https://www.youtube.com/shorts/00Hck2_hbJU)
Video language: English

[Complete guide](https://lukejbyrne.com/library/guides/olpw4e0fp2s/index.md)

## Continue with the complete guide

LFM3 Omni from Liquid AI takes a picture, a recording and a text prompt together. We ran it locally on a Mac with a supported llama.cpp runner, the Omni model and its matching projector; the two downloads were about 1.6 GB, plus the memory needed to run them. It answered our chart-and-recording question correctly, but in a separate chart question it got the total number of people wrong and only corrected it after feedback.

The complete guide has the starter files, the exact command, the test results and their limits.

## The takeaway

Check figures against the source image: a correction after feedback is useful, but you want the right answer the first time.

## Workflow resources

- [huggingface.co](https://huggingface.co/LiquidAI/LFM3-Omni-2B-GGUF)

## Sources and limitations

Short companion prepared 4 November 2026 from the approved English Short and the saved 2 November test files. The linked full guide covers the setup and test in detail.

[Original video](https://www.youtube.com/shorts/00Hck2_hbJU)