The general producer of the hackathon was Irakli Tsarukyan. In the AI production, Denis Mishchenko from SKILLOUT and VidMK participated. And Nikita Stepanov, the AI creator, turned our ideas into a script during the discussion.
First, Nikita fixed the mentor's appearance so that the audience would recognize the same person in each scene. Then he selected location references, generated scenes using GROM, and in Workflow, he combined them with the narrator's voice and music. The inscription on the airship couldn't be generated — it was added during editing. The finished video in the show recording, 2:48:16.
They named the village 'Poludenniy.' On its website, you can browse a resident's day from 6 AM to 11 PM. At the request of the hosts, we added voting for the audience during the broadcast. And at the end of the page, you can tell about yourself and get your day's schedule in the village. The site "Halfday".
What's next
By November, we plan to use the voice GROM again at offline events. By that time, we'll add a file queue, prepare a name dictionary, and check how the system matches a person's quotes in adjacent fragments.
Voice GROM on devices is a separate direction we want to develop. We need to figure out on which devices it can work and in what situations people will find it convenient to use it via voice.
For this, we need not only money but also video cards, other computational resources, and the devices themselves on which we will test the assistant. Currently, we are discussing this with investors.
Also, in GPTunneL, we want to link the voice assistant with Workflow. During the hackathon, Nikita manually transferred our solutions into the script. I want that next time he could open the project and say: "GROM, redo the evening scene." And the assistant would already know from our conversation at the table where we decided to move the festival and why.
I write on Habr infrequently: writing articles takes a lot of time, and it's never enough. I often talk about the development of our products in my Telegram channel.
