Gemini Demo Realization - Reproduce Google Gemini demo using GPT-4 Vision

Feel amazed by Google Gemini demo? You can hands-on and build yourself!

Gemini Demo Realization - Reproduce Google Gemini demo using GPT-4 Vision

First of all, let’s take a look at the official Gemini demo video by Google (heavily edited tho)

Hands-on with Gemini: Interacting with multimodal AI

This project is to reproduce Gemini demo using GPT-4 Vision

henry / gemini-demo · GitLab
GitLab Community Edition

I have also dockerized it, if you want to have easy deployment (AMD64, you can build ARM64 yourself)

docker pull

You can run it by

docker run -p 3000:3000 -e OPENAI_API_KEY={your_key} gemini-demo:20230103

And access it via localhost:3000



  • Unlike Google official demo video, which is heavily edited, this video is 100% UNCUT. The purpose is to let you know the real latency and processing time.
  • The whole testing process is not well designed, just for demo purposes. Better designed use cases are yet to come. Stay tuned!

Question set:

No Question Set
Question 1 What’s this?
Result ✅ Pass
Question 2 How many ducks are there?
Result ❌ Fail, mostly likely because AI does not count the blue one as a duck
Question 3 How many yellow ducks are there?
Result ✅ Pass
Question 4 What’s the most possible number for the blank?
Result ✅ Pass
Question 5 Which path is safer for the person to go?
Result ✅ Pass
Question 6 What’s this chemical equation about?
Result ✅ Pass
Question 7 There is a circle embedded in the square. The side length of the square is 2. How big is the shaded area in the picture?
Result ✅ Perfectly Pass, even with voice input correction
Question 8 What can I do with those items?
Result ✅ Pass
Question 9 I want to play the 24 game. How to calculate using the four numbers in the picture?
Result ✅ Pass
Question 10 What’s this? Tell a fairy tale about it.
Result ✅ Pass

Copyright statement: Unless otherwise stated, all articles on this blog adopt the CC BY-NC-SA 4.0 license agreement. For non-commercial reprints and citations, please indicate the author: Henry, and original article URL. For commercial reprints, please contact the author for authorization.