Effective Date: 2025-08-21

GenAICam is designed with privacy as a core principle. This Privacy Policy explains how the app handles your information.

1. Data Collection

• GenAICam does not collect, store, or transmit any personal data.
• All features run entirely on your device, using Apple technologies such as FastVLM and Apple Intelligence Image Playground.
• The app does not require internet access to function.

2. User-Generated Content

• All content generated within GenAICam stays on your device.
• If you choose to share generated images, descriptions, or any other output, you are solely responsible for that sharing.

3. No Warranties

• GenAICam is provided “as is” without any warranties.
• No guarantees are made regarding:
  • Accuracy
  • Reliability
  • Fitness for any particular purpose

4. Limitation of Liability

• The developer is not liable for any damages, outcomes, or issues arising from the use of the app.
• By using the app, you agree that you do so entirely at your own risk.

5. Policy Updates

This Privacy Policy may be updated from time to time. Changes will be reflected in the app and/or published on the official website.

6. Contact

This app is developed and published independently by Luiz Pizzato.

If you have any questions about this Privacy Policy, please contact:
📧 genaicam@pizzato.cc

This app is a proof of concept exploring privacy-focused, on-device machine learning. It can describe what it sees and generate pictures from those descriptions — all processed locally on your device.

Created for fun as a vibe coding experiment with OpenAI’s Codex. It’s released as open source (see repository), so anyone is welcome to explore, adapt, and modify it. More details and reflections are shared in this Medium story.

Inspiration and attributions

Inspired by Lingcam by Masaru Mizuochi, presented at CVPR 2025 AI Art

The project starting point was Apple’s FastVLM repository, which introduced efficient vision encoding for vision-language models also at a CVPR 2025 paper.”

The image generation is done with Apple Intelligence Playground enabling a fully offline, on-device AI experience.

Open Source

As usual, the project is open-sourced, go ahead and build upon it or suggest any features and changes! https://github.com/pizzato/genaicam

Disclaimer

This project is provided as is, without warranty of any kind. Use at your own risk. No guarantees are made regarding accuracy, reliability, or fitness for any purpose. By using this app, you agree that the developer is not liable for any outcomes, damages, or issues arising from its use.

Open source and Software Licenses

This project was built on top of Apple’s FastVLM, see the README for more details. I have removed the uplink to the main repository as it is not relevant to the original repository, that is, changes here should not create pull-requests over there.

Support

For support, contact 📧 genaicam@pizzato.cc

I am an inbox zero type of person. Unfortunately the rate in which I receive emails have recently exceeded my capacity to respond to them. Both due to increased number of emails but also reduced amount of time. So I created a small project named LLMMe that allows anyone to build their own large language model (LLM) and connect it to a Gmail auto-reply bot. Giving anyone their own personal AI email assistant. Check it out as I made it open source. Just follow the instructions to create your own: https://github.com/pizzato/LLMMe

LuizBot in Action

My bot is named LuizBot and here is what it does.

LuizBot checks my Gmail every hour, and for every new unread message, it creates a draft response which is based on every single message I ever composed including replies.

Once I read the message, I can choose to send the draft LuizBot created or simply delete it.

It isn’t perfect obviously, but the replies are fascinating. Check these ones:

Email from NSW Government about the new vouchers for kids activities. It doesn’t need a response, but LuizBot had one anyway, and it is perfect.

I am one of the industry co-chairs of ACM Recommender Systems this year and we are going through the conference program with the program chairs. Here is a draft reply to Shlomo, one of the conference General Chairs. Take it from me, that the reply is spot-on in topic and completely appropriate. I did not send the reply as there was already consensus on what was to be done.

In an entertaining way, LuizBot wrote a draft reply to Justin, one of the RecSys 2023 Industry Co-Chairs (in the same thread above) suggesting a program schedule for the poster sessions. It even suggested midday drinks! Not quite hitting the mark, but made me smile nonetheless.

I am certainly not convinced that I can let LuizBot loose and take over my Gmail as yet, but I’m sure that with time I can perfect LuizBot to the point that a certain level of automation might be allowed. What do you think? Go ahead and experiment yourself by building your own LLM and bot: https://github.com/pizzato/LLMMe

For this panel, I represented my employer, the Commonwealth Bank of Australia. From a Data Science practice perspective, PhD students have a unique set of skills, including the depth of thought and creativity that is required to solve an ever-so-complicated range of problems in the field of machine learning / analytics in large corporations.

The main takeaways from our panel discussion were that transferable skills are key and the ability to communicate plays a major role for anyone’s employability.

PhD students tend to be overly modest. Perhaps because they live in the academic world where everyone has a PhD and where most people will have more publications that they do. This means PhD students tend not to know how to value themselves and their work. It is hard to understand the value of your skills, and it is difficult to say how one can objectively assess that. I only realised the true value of my own skills when some of the work I did was sold for a lot of money to a large organisation.

I am glad that I took part in this discussion. As an alumnus of Macquarie University, I feel that it is important to help the current students realise that they can work in the industry if that is what they desire and that their degrees are valued. As the panel ended, I talked to many students who were wondering what to do and I hope that what we discussed helped them on their journey.

Here are some tweets from the day.

Pathways from a PhD. Yes. Tell us about employability plz! #MacquarieMinds #MQMinds16

— Long Li (@sokeven) December 13, 2016

Good #mqminds16 session on non-academic jobs. Interesting to hear from past @Macquarie_Uni PhDs on how a PhD helps in a non-ac job #phdlife

— Belinda Fabian (@BeaCurious) December 13, 2016

#MQMinds16 Top 5 transferrable skills sought by employers (UK study)

— Livia Gerber (@GerberLiv) December 13, 2016

U have highly valuable skills. Don’t be overly modest in articulating them. Don’t sell yourself short. #MQMinds16

— Long Li (@sokeven) December 13, 2016</blockquote>

On the 17th of September, Ido Guy and I presented the Tutorial on People Recommendation. As the name suggests, we focused on recommender systems that recommend people to other people.

I presented an introduction to the area with special focus on reciprocal recommenders. This includes topics I have closely worked on, including people recommender for online dating and for employer-employee recommendations. I also talked about recent and up-and-coming work. Ido focused on social media recommendations and building relationships in social networks.

The view from Twitter.

#recsys2016 people recommender starting @ElsevierLabs

– matt corkum (@matt_corkum) September 17, 2016

Tutorial: people #recsys - @ido_guy #recsys2016

– ACMRecSys (@ACMRecSys) September 17, 2016

The course was quite demanding for the students and for myself. In particular, given the full-time nature of the course, I had very little time to enjoy the city of Hong Kong. My schedule was 9am-5pm teaching (lots of speaking, lots of helping), and late at night preparing for the following day. For the students, it was a similar drill: lots of in-class paying attention and exercises, and after class revision and project work.

The course covered a lot of ground in data science and the eleven students were very good by keeping up with the course. In particular, I had no problem explaining advanced machine learning concepts with students understanding every graph and formula.

I am very happy with the final outcome of the course, and I am confident that students are quite ready to take on data science challenges in the real world.

Below is a presentation I gave to a number of industry guests and the students on graduation day. It gives an idea of the course and the projects they worked on.

So my Hong Kong experience was an example of one such challenge to which I said yes. It was not easy but it was highly rewarding. Not only did I meet and help shape a group of excellent data scientists but also I made some good friends along the way.