I originally wrote ‘I Got Voted Off the Island’ as a long-form Medium article about the personal side of an extended executive job search. I held off publishing it for a while, for many of the same reasons the article itself explores.
As an experiment, I used AI-assisted songwriting and music generation to turn the article into two very different songs. I then used ChatGPT, an MCP connection to RunwayML, and AI video generation to create a music video for the lighter one.
A longer, more reflective interpretation of the article and the uncertainty of looking for what comes next.
A shorter, lighter country interpretation that pokes fun at the repetition of the executive interview process.
The full AI-generated music video for the lighter, country interpretation.
I used ChatGPT to brainstorm and iterate on two sets of song lyrics: a longer, more serious song and a shorter, more lighthearted song. I then used Suno to generate both songs, experimenting with different musical styles and model versions.
For the lighter song, I worked with ChatGPT to develop and repeatedly refine a scene-by-scene screenplay for a full music video. Using a connected MCP integration, ChatGPT submitted prompts to RunwayML and helped drive the creation of 25 short video clips, generally eight seconds each.
I assembled the generated clips with the finished song in iMovie, adjusted the sequence and timing, and exported the complete video as an MP4. Finally, I used FFmpeg to reduce the file size for web hosting.
Published the long-form article that became the source material for both songs.
Helped brainstorm and iterate on two sets of song lyrics, then developed and refined the screenplay and individual prompts for the music video.
Generated both songs, with iterations across different musical styles and model versions.
Connected ChatGPT to RunwayML so it could submit generation prompts and help manage the clip-creation workflow.
Generated the 25 individual video clips used to construct the music video.
Combined the finished song and generated clips, handled the final sequencing and timing, and exported the completed video as an MP4.
Compressed the exported MP4 to make it practical for web hosting.
One lesson was how differently the same source material could be interpreted. The reflective and lighthearted songs began with the same article, but choices about lyrics, structure, musical style, and pacing produced very different results.
Creating the video also reinforced that longer-form AI video is not a single-prompt exercise. Building a three-and-a-half-minute video out of roughly 25 eight-second clips meant that continuity details didn’t automatically carry over from one clip to the next. Details like characters, clothing, locations, and props had to be actively remembered and re-supplied for every new clip. You can actually spot a few places in the video where I missed one, like a volleyball net that switches between yellow and blue between shots.
The MCP connection reduced some of the friction involved in moving between the screenplay and video generation, but it did not remove the need for iteration and judgment. The final story still depended on selecting usable clips, arranging them against the music, adjusting the timing, and deciding what best supported the song.
I also noticed that ChatGPT would sometimes default to its own built-in image processing instead of using the RunwayML MCP connection, even after being prompted to use it. I often had to remind it partway through a task to route the request through RunwayML instead.