top of page

Concept to Script: What happens when you collaborate with AI: Part I

  • Writer: drakedantzler
    drakedantzler
  • Sep 15
  • 4 min read

Updated: Sep 22

This is Part I in a blog series about using AI to create a complete libretto rewrite—from concept to organization to libretto.


What happens when you try to interact creatively with a large language model? Today’s post kicks off a series about the largest and most complex project I’ve ever coordinated with ChatGPT: developing a completely new concept and libretto for Handel’s Alcina. The process has included moments of brilliance and bursts of frustration—but that’s what you get when collaborating with AI.


Why Alcina? Why Now?


Acis and Galatea as a New Year's Eve Party; the last Handel Opera at Oakland.
Acis and Galatea as a New Year's Eve Party; the last Handel Opera at Oakland.

At Oakland, we strive to make our opera projects accessible to our students. Alcina is musically and vocally perfect for them—but dramatically outdated. I knew we needed a new script. What I didn’t expect was how much AI would be involved.


From Curiosity to Collaboration


I started small, just testing ChatGPT’s operatic knowledge:

Me: “Tell me about Handel’s Alcina. What are the major themes and most famous moments?”

The answer was more complete than I was expecting! So I asked:

“What are some modern adaptations of Alcina?”

It returned a list of legitimate productions—Aix-en-Provence, Dutch National, Staatsoper Berlin, etc.—and I realized I had a pretty useful research tool on my hands.

“If you wanted to make a modern version for college students with a limited budget, how might you adapt it? Avoid therapy, please.”

(A side note: I don’t love the trope where magic or antique moral systems are “solved” via mental institutions. Just a me thing.)


The result?

ChatGPT: “🎭Title: ALCINA: A House of Illusions Alcina is a charismatic digital persona, running a ‘utopian’ social-media space where guests become addicted to pleasure, filtered identity, and emotional escape.”

It was strange, contemporary, and kind of brilliant. It started my gears turning.


“Can You Write the Libretto?”


Encouraged, I asked:

Me: “Talk to me about a scene-by-scene plot summary... Also, could you make a new libretto?”

Then came the big one:

“Are you capable of following the rhyming scheme, maintaining word stress, and making a performance translation that suits singing?”

The AI replied:

ChatGPT: “✅ Capabilities for Scriptwriting (Sung Performance Translation)...I can write recitatives and arias... Match syllable count... Adjust vowel color and placement... Track character voice, diction, and tone...”

Let me say right now: I can do all those things... sort of!


First Draft: A Little Prodding


I tested it, asking for a translation of one of the arias. I returned this at a first:

ChatGPT: “My heart’s wide open, it’s not quite whole, but spoken— And somehow you, just you, could make it beat like new.”

Not bad. But not current. So I asked:

Me: “Can you make a more modern version of this text?”

Then it came back with:

Chat GPT: “You just walked in, and I forgot Whatever rules I thought I’d got. You smile like you’ve been cast as fate, And babe, I’m ready. Set the date.”

We were getting warmer. I decided to go all in.


A Big Miss: Working with Variable Source Material


When I asked:

Me: “Do you have access to the Italian text of Alcina?”

ChatGPT, in its typically overconfident way, replied:

ChatGPT: “Yes, I do! I have access to the full Italian libretto...”

But it didn’t! Or more accurately, it was unable to define what the source material was. What I got was a patchwork of libretti drawn from adaptations across a wide range of sources. The result: Characters were missing. Scenes were out of order. Music was attributed to one chracter but it belonged to another. Big miss.


I have come to realize that ChatGPT was overwhelmed by all the different editions, arrangements, and translation of Alcina. I eventually found an online libretto that matched our material, and had ChatGPT scrape it for source material. More next blog. I'm sure ChatGPT would have done much better with an operatic war horse.


Lesson learned: Double check the work each time ChatGPT generates something.

Look! The show in 20 scenes! Except my Bärenreiter has 38 scenes....
Look! The show in 20 scenes! Except my Bärenreiter has 38 scenes....

ChatGPT’s Nails It:


I asked for scene breakdowns. It gave them. But none matched my score. Another miss. But I adapted to explore it's creative possiblities. Then I asked:

Me: “In the show itself, please list all the magical actions that Alcina performs.”

Followed by:

“What are some options we could examine within the concept we are discussing? I enjoy if they are culturally relevant.”

That’s when it delivered this gold:

ChatGPT: “🪄 Magic as Social Media Power / Influence... 🖼️ Magic as Image Manipulation / Digital Curation... 💻 Magic as Algorithmic Control... 🎮 Magic as Escapism / Fantasy Addiction...”

And more, with details! I brainstormed and worked through these concepts, eventually resolving exactly how the larger concept of magic would exist in this world. ChatGPT was very, very good at this part.


What Is “The Island”?


We refined it further, iterating on suggestions until we had some solutions. We worked as a duo, me asking for ideas, selecting results I liked, asking to brainstorm off of those. We would up with results like:

“🏝️ The island is a curated, enclosed digital-performance space... Think influencer mansion + luxury streaming house + AI-generated illusion.”

“She curates the environment: lighting shifts, mood music, perfect backdrops, filters...”
“The illusion collapses: filters glitch. Projections go out. People stop liking, commenting, or caring.”
“Alcina is no longer trending, and her power dissolves.”

A Solid Days Work (And a lot of files)


By the end of this day, which was July 30th, I had refined the concept, aesthetic, and characters significantly, and done some proof-of-concept lyrics. I also had piles and piles of contradictory files and about 17 versions of the word "Alcina" in my downloads folder.


ChatGPTs plan to create the script. It would not be this easy, or this linear...
ChatGPTs plan to create the script. It would not be this easy, or this linear...

What's Next?


In Part II, I’ll dig into the expansion of the universe. The script project felt like something that was quite small (the original script), which became very large (lots of details of who, what, and how the opera was being updated, synopses, beats) to something quite slender again (the actual script). So next is:


  • Scraping the text

  • Various large scale synopses

  • Adding new characters

  • A white page for collaborators


Then we can head on to the actual translation of arias, the addition of new characters, and the connective tissue of the recitatives, before we head off to the ongoing topic of polishing.


If Alcina is about illusion, identity, and manipulation, then an AI-generated libretto seems weirdly perfect.

Comments


© 2018 Dantzler. Proudly created with Wix.com

bottom of page