Chica de Rojo - "Te Siento"

A Latin pop song by fictional artist "Chica de Rojo", from her album "Amor Inmortal" (2025).

Comment

Inspiration

Once upon a time, in my 20s, I used to play in garage rock bands in my local city, even recording a few albums in studios. Great times in my life that I sometimes miss in my 40s. I've been using Udio for roughly the past year to make full length radio tracks, just for fun, from all music genres. I generated quite a few songs lately, intending to use the best couple of the bunch (in my opinion) for this hackathon. I had never before used AI to generate video, so this was an excellent chance to give that a try. I chose this song to use because it was catchy, pleasant, and truly gave me "real song" vibes.

What it does

Mimics the sound of a dark-club Latin Pop song

How we built it

The song was made using Udio. First, an artist style-reference song was loaded in (for this song, and every song on the album, that style reference song was "Siempre al Borde", the first track on the album). This ensured that the singing artist was the same, and sang each song in a similar style. The prompt for the entire album was: Dark, Latin Heritage, Bass-house, while the unique prompt for this individual track was "Missing the one person you love but cannot see or touch". After a satisfying 30 second snippet was generated (in this case, the verse & chorus), it was remixed to make the singing clearer, and to properly frame and tag the lyrics ([verse], [pre-chorus], [chorus], etc), and to make any small edits to words or phrasing. Afterwards, the rest of the song is built out, first generating a second [Verse], then building the song out to the end, then finishing the song with its intro. When generating music, my mantra is "repetition, repetition, repetition". To get one 30 second section of song (or less) that sounds "right", hundreds must be generated.

The video clips were made using a mixture of Meta AI & Hailuo (photo to video). I used Meta AI to generate photos from text (prompt: "a pretty Spanish girl with red hair"), then selecting one of those photos to be a style reference for the AI. That photo was used to generate many more photos of that same girl, wearing different outfits and in different settings. A small number of those photos (approximately 4) were animated using Meta AI. The remaining (dozens) were fed through Hailuo to create the bulk of the video clips used in the video. Eleven of those video clips were imported into Vozo AI alongside chosen snippets from the song to lip sync the artist to the music. Again, repetition wins the day. Some scenes only took a handful of generations to get what I was looking for. However, most scenes took many many iterations to get it "right". In all, for the video, roughly 90 individual video clips were generated, ranging from 1 second to 8 seconds.

The sound FX for the initial opening 6 seconds of video (city night-sounds, nightclub line-crowd sounds, car and car-door sounds) were created using ElevenLabs AI (text to sound).

The music video, combining all the above, was produced by me using Microsoft Clip Champ. Clip Champ is super easy to use, and fun to use as well, and provided all the tools I needed to assemble and edit the video.

Challenges we ran into

I've never, prior to this hackathon, made AI video (though I have good experience using AI to generate images, music, sounds & speech, 3D models, and Typescripting). Making video with AI was more tedious than other things, sometimes frustrating.

Accomplishments that we're proud of

This is my second attempt at an AI vid, my first at trying to lip sync music and create a realistic feel

What we learned

I learned a good bit about editing short videos, and generating AI videos.

What's next for Chica de Rojo - "Te Siento"

Chica de Rojo "Amor Inmortal" officially releases on all streaming platforms November 19, 2025

Built With

Share this project:

Updates

Submission history