Keynote at ADOS Paris

On 28 March 2025, I attended the amazing ADOS event in Paris at Artifex Lab, a new space in Paris for bringing together AI arists and technologists. ADOS was co-organized by Banadoco and Lightricks, which was an ideal fit for Artifex Lab: Banadoco is the leading online community for AI art+technology and Lightricks’s LTX is the fastest generative AI video model (and it’s Open Source on top of that!).

Coincidentally, our team has recently been investigating LTX deeply and also has been hanging out a lot on Banadoco’s inspiring Discord server. One thing led to another and I was invited to present a keynote at ADOS. We brought most of the team, as well as a bleeding edge demo. Unfortuntately, I missed out on the second day of the event, which featured a hackathon, where our PhD student Hovhannes received a prize (see his upcoming blogpost on our team webpage).

Before reporting about the keynote and demo, I would like to highlight some of my favourite things that I saw at the event. All talks are recorded on youtube; I provide direct links to the talks in the following.

I enjoyed Emma Catnip’s talk, particularly her work Bloom. It is very soulful and beautifully designed. I will never forget her bitter-sweet explanation: “It’s about the summer of love I that never had”.

Another very interesting artist’s talk was by Vibeke Bertelsen (Udart); timestamp. Imagine current AI tools in the hands of H.R. Giger… Below is one of her award-winning short movies. Definitely most unique!

On the tech side, my favourite talk was by Yvann Barbot. In his talk, he showed how easy it is to generate cool audio-reactive video clips using his ComfyUI nodes. One example below by Visual Frisson. After the event, I checked Yvann’s youtube channel. I can highly recommend it!

Finally, let me get to my keynote talk and our demo.

Due to the extreme technical difficulties involved in our demo, we were working literally till the last minute on it. Chapeau to Dávid Maruscsák and Francesco Dettori for their very hard work near demo time! Very unfortunately, the backend of our demo stopped working 5 minutes before my talk started. As a result, we could not show the demo during my keynote (as we had originally planned), but only 2 hours later. Luckily, most attendants of ADOS were still there and could give it a try!

We demonstrate how to use a cloud backend for highspeed generation of spatial videos based on image and text inputs. You can see a similar version of the demo below; however, compared to the demo we showed at ADOS, it is only using 1 RTX4090 GPU for inference, as opposed to 4 H100 GPUs at ADOS, so it runs significantly slower.

You can see my keynote on youtube, where I speak about my long past with AR and my more recent past with AI, and how everything converged into this demo :-)

To conclude, I would like to sincerely thank:

• The ADOS organizational team: Banadoco, Lightricks, and Artifex Lab.
• Special thanks to Lightricks for giving us early access to their distilled LTX model, which enabled impressive speedups.
• All ARAI team members for bearing with my demands :-) Directly involved in the demo were Cyrille Leroux, Thomas Betton, Dávid Maruscsák, Francesco Dettori, and Hovhannes Margaryan.