Shunned at a Funeral

Shunned at a Funeral is a virtual band created by Ethan Howard, an Oregon Tech assistant professor and lifelong guitarist who built the project to continue making art despite a progressive visual impairment. The result is one of the most thoroughly documented human-AI creative collaborations in contemporary music — and an outstanding case study for any CS class discussing prompt engineering, generative models, and digital ethics.

The three female performers fronting the band in videos and promotional imagery do not exist. They are entirely AI-generated characters. The music videos and cinematic tie-ins — including feature-length AI films Bethlehem and Moriah — are created entirely using text-to-video and text-to-image tools. The lyrics, however, are entirely human-written: Howard composes every word by hand, drawing heavily from biblical scripture and Reformed theology.

Howard’s production process is a multi-step iterative pipeline worth diagramming in any CS class:

1. The Seed. Howard records a base rhythm seed — sometimes literally tapping on a coffee can — to anchor the time signature before AI enters the picture. Human input grounds what follows.
2. The Generation. The rhythm seed, paired with highly specific structural prompts (demanding particular harmonies, heavy metal guitars, or Celtic melodies), is fed into Suno, an AI music-generation platform.
3. The Refining Loop. Rather than accepting one output, the track is generated dozens of times. Audio stems (vocals, drums, guitars) are extracted, imported into Audacity, manually cut and remastered, and recursively fed back into the AI. The final track is assembled from the best fragments across many generations.

This is not “push a button and get a song.” It is a software pipeline where the output of one model becomes the evaluated input to the next — exactly the kind of multi-step algorithmic workflow students will encounter in real engineering careers.

1 — Spot the AI Artifacts. In the live performance clips, classic text-to-video failures are visible: a guitar neck’s fret markers shift arbitrarily mid-video, and the headstock morphs shape between frames. Ask students: why does generative video struggle with object permanence? This opens a discussion on how video diffusion models generate frames without a persistent 3D spatial model.

2 — Algorithmic Iteration vs. a Single Prompt. The Suno-to-Audacity loop illustrates that modern AI engineering rarely involves a single prompt. It involves pipelines, evaluation criteria, human judgment about what is “good enough,” and iteration. The AI is just a new kind of function call inside a larger human-designed workflow.

3 — The Ethics of Disclosure. The band originally faced significant online backlash because their marketing presented the AI characters as real musicians from Oregon. This is a documented case of the deepfake disclosure debate: at what point does creative fiction become deceptive? What obligations do creators have to audiences when using generative AI for identities and personas?