GitHub radar
An AI model that sees, hears, and talks back — right in your browser
LiquidAI released a tiny AI model that takes text, photos, video, and voice as input and answers with text or speech — you run it in your browser in about ten seconds, no install or API keys needed.
It's a single ('omni') AI model with only 2.2 billion parameters (internal settings) that merges separate text and image models into one — built specifically for phones and laptops.
When it helps
Good for quick experiments: ask a voice question about a picture or video and get an instant answer. For serious voice and video apps, look at bigger models.
Pros
- Runs straight in your browser — nothing to download
- Loads in about 10 seconds
- Takes text, photos, video, and voice at once
- Understands more than twenty languages
Cons
- It's a tiny model — noticeably weaker than big AI models
- Not built for serious voice or video projects
How to set it up — step by step
- 1Open a recent Chrome or Edge browser — you need WebGPU support.
- 2Go to huggingface.co/spaces/LiquidAI/LFM3-Omni-2B-WebGPU and wait for the model to load (about 10 seconds).
- 3Send the agent the text from the block below so it explains what the model does and how to use it.
- 4Upload a photo or video and ask your question by typing or speaking.
- 5Check the answer — the model replies in text or speech, in one of over twenty languages.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open this space https://huggingface.co/spaces/LiquidAI/LFM3-Omni-2B-WebGPU and explain what this model can do and how to use it: what you can feed it (a photo, video, voice) and how to get an answer back in text or speech.
Nothing to install — it runs right in the browser via WebGPU; you need a modern browser (Chrome or Edge, for example) that supports it.
Open on Hugging Face▌ More finds