Gemma 3n First Look & LOCAL Chatbot Demo (TTS, Vision, Roleplay)

Timestamps:

00:00 - Intro
01:26 - Technical First Look
03:48 - Giving It A Voice
04:17 - React App Chatbot Demo
07:18 - GGUF Mentions
07:57 - Multimodal Testing
09:20 - Thoughts RE Omni Models
10:45 - Real Image Testing
12:45 - Refusal Testing
14:37 - Roleplay Testing
16:00 - Translation Testing
16:54 - More RP Testing
18:40 - Script Overview
25:45 - Offline TTS Tips
26:14 - Closing Thoughts

AI Consulting: https://bijanbowen.com
Discord:   / discord  

In this video, we take a look at Gemma3n, a newly released native multimodal model from Google DeepMind. Originally built for edge devices like phones, it’s now available on Hugging Face and can be run locally on a variety of systems using multiple backends.

We start with a brief technical overview of the model, then show how to integrate it with TTS (text-to-speech) to give it a voice — allowing the model to listen, respond, and speak, fully offline. We demo a simple chatbot using a React front end and walk through speech pipeline scripts, covering both the GGUF mention and the backend options.

In the second half, we explore the multimodal image capabilities, test image understanding, translation, refusal handling, and even a few rounds of roleplay to see how the model behaves with character prompts.

HF Repo: https://huggingface.co/google/gemma-3...

Github For React App: https://github.com/OminousIndustries/...