Building the Map: How ChatGPT Helped Reimagine Alcina: Part II

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

Where I left off...

Before translating a single line of Handel’s Alcina, I had to build a map: a massive hybrid of 18th-century drama and 21st-century influencer culture. That meant organizing the entire opera—arias, recits, and choruses—not just as historical text but as a living script for our adaptation.

By this stage in the project, I had made significant progress clarifying the adaptation plan. Still, before I could dive into translating text, there was a mountain of organizing, research, and detail work that needed to be tackled.

Last week, I described how the project felt like it ballooned outward before becoming “slender” again. This was the ballooning.

Pulling the Aria Texts

The first hurdle was getting a clean copy of the libretto. If I asked ChatGPT to fetch it on its own, the result was always a garbled mess. The only way forward was to give an exact command:

“Scrape, using your tool”

—and point it to the specific source: FlaminioOnline - Handel's Alcina libretto. If I didn’t use those precise words, ChatGPT would pull in the wrong text.

Once we cracked that code, things moved quickly. ChatGPT generated a list of every aria with a concise description of the dramatic action. For example, here is part of what was generated:

• O s’apre al riso – Morgana; Morgana flirts openly with Bradamante (disguised as Ricciardo), captivated by “his” appearance, and expresses her instant attraction.

• Di’, cor mio, quanto t’amai – Alcina; Alcina invites Ruggiero to show the newcomers the magical places where she and he once fell in love, recalling those moments as proof of her devotion.

• Chi m’insegna il caro padre – Oberto; Oberto laments his missing father (in my version, brother), begging anyone to help him find him and expressing his despair.

• Di te mi rido – Ruggiero; Ruggiero mocks Bradamante and Melisso for warning him about Alcina, declaring that he is happy to follow love and beauty rather than honor or reason.

• È gelosia – Bradamante; Bradamante confronts Oronte and Morgana, accusing them of jealousy and explaining how it torments the heart.

  
  

Reframing the Action

With a summary created, it was time to push on to adaptation. Next, I asked ChatGPT to propose modernized versions of the action for each aria, within the world we were building. The result was a table that compared original dramatic intent with two possible influencer-era interpretations.

For example:

• O s’apre al riso – Morgana

  • Option 1: Morgana is charmed by a new "conent creator" joining the house, filming a TikTok duet together.

  • Option 2: Morgana treats Bradamante like a breakout celebrity guest on a livestream, showering them with attention.

• Chi m’insegna il caro padre – Oberto

  • Option 1: Oberto searches the house archives, begging for help finding a lost sibling from before the influencer life.

  • Option 2: Oberto posts a heartfelt story asking followers if they’ve seen his brother in the curated online world.

    
    

As I mentioned last post, ChatGPT was generating numerous files. Those external files were almost always poorly formatted and, frankly, wasted time. By this point in the process, I moved to asking ChatGPT to make everything “inside the chat” rather than creating new documents. I strongly recommend modifying your prompts to create tables and output inside the chat, and only ask for external files when absolutely necessary.

We repeated the process for the entire script—arias, choruses, and recitatives—so I ended up with a complete map of the show: the original dramatic structure, plus two modernized versions of each scene. What fascinated me here wasn’t just the accuracy—it was how easily the opera’s emotional beats translated into influencer dynamics.

Act III Sample

Here’s a snapshot from Act III to give a sense of how rich this mapping became:

• Credete al mio dolore (Morgana) -

  • Original: Pleads with Oronte to believe her love.

  • Modern: A teary apology livestream, or an “open letter” Instagram carousel defending herself.

• Mi restano le lagrime (Alcina) -

  • Original: Only tears remain; she wishes for escape.

  • Modern: A stripped-back no-makeup video, or a voiceover montage of empty rooms.

• Barbara! io ben lo so (Oberto)

  • Original: Recognizes transformed father/brother and vows her downfall.

  • Modern: A confrontation livestream exposing secrets, or uploading proof of transformation tricks.

    
    

With this full map in hand, the next challenge was figuring out how to stitch everything into a coherent synopsis.

The Synopsis Problem - Making sure the big picture worked

At this point, I wanted to integrate everything into a synopsis. I was getting concerned that my expansions would not synthesize into a coherent show. But here’s where I hit a common frustration: when I asked ChatGPT to pull up a synopsis we had already created, it simply made a new one on the spot. Similar, yes, but not the same—and not what I needed.

That “generic” memory meant I had to dig up the original from another chat and reinsert it to keep things accurate. As you go, you have to prompt ChatGPT to only change what you explicitly need, in particular if you are recreating large quantities of text.

Once that was solved, I tried a new angle:

Make three detailed versions of the synopsis, based on choosing different options from the table above. Each should be logical, continuous, and plot-conscious.

3 Narratives to choose from.docx

Download DOCX • 18KB

This was a turning point. ChatGPT is very good at complete "choose-your-own-adventure" tasks. ChatGPT instantly producing three different longform, coherent, logical synopses. After choosing my favorite of the three, I repeated the process to explore three separate Act III endings. Once again, easy-peasy for ChatGPT. Soon I had a 3,500-word narrative draft. I went back and created a bolded trail marking which original options ChatGPT had selected from my choices table. Now I was armed with both a narrative and a structured table to reference individual moments. You can click on the link to see one of the steps.

Adding a Cut List & New Characters

With complete narratives and individual scene charts, it was time to integrate our cuts. Alcina is quite long, and we had drafted some cuts to get the show down to about 2 hours or so. This was easy enough, with ChatGPT making some slight modifications to the narrative. From there, we moved into a signature Oakland tradition: adding new characters to expand casting opportunities.

I asked ChatGPT to brainstorm 20 possible cameo moments for new “influencer chorus” members—short solos or duet fragments, either B sections or da capos of arias in the show. The prompt asked for moments that could deepen or comment upon the themes we were working with. It came back with a detailed act-by-act list, including functions like:

• “Influencers hype Alcina’s mansion, echoing the main chorus’ warning but twisting it.”

• “An influencer provides a ‘testimonial’ about Alcina’s allure.”

• “Influencer cameo contrasts Ruggiero’s regret in ‘Verdi prati’—they’re still clinging to the beauty.”

• “Influencer delivers a chilling cameo—turning Alcina’s ruin into content, recording her breakdown.”

This gave us an elegant way to insert new voices without disturbing the larger architecture of Handel’s opera.

Refinement & Sharing

From there, it was a matter of refinement. The 3,500-word narrative was trimmed into a one-page version I could share with my collaborators at an upcoming meeting. That version, below, became the working draft for us to test cuts, stage choices, and character additions:

Alcina Concept Summary – Collaborative White Page

This document outlines the core concept for our upcoming production of Handel’s Alcina. It is intended as a collaborative white page to spark design, staging, and rehearsal conversations. The production reimagines Alcina’s island as a curated influencer mansion where illusion is created through projection, emotional control, and digital performance aesthetics.

Concept Overview

Alcina’s island is a hyper-stylized content mansion—a luxurious, curated space designed to project an idealized fantasy. This space includes real rooms (lounges, a stage, confessionals, and a control booth) but is layered with illusion via lighting, projection, and social media metaphors. Magic in this production is interpreted as Alcina’s control over her narrative, appearance, and environment, similar to a high-powered influencer managing a viral brand. When her control fades, the house glitches and collapses, revealing emotional vulnerability and aesthetic fragility.

The Space and Its Magic

• The house is a blend of physical space and digital illusion.

• Projection and lighting design are critical to representing 'magic': filter overlays, algorithmic mood shifts, and aesthetic control.

• Characters inside the house live within this illusion; they do not acknowledge the audience.

• Alcina’s power manifests visually—when she sings, the space responds (lighting shifts, camera tones, glow effects).

Key Characters and Dramaturgical Roles

• Alcina – The curator and controller of the illusion. Her downfall is both emotional and aesthetic.

• Morgana – A performer within the system, addicted to attention and image.

• Ruggiero – A collaborator drawn in by the fantasy of perfection.

• Bradamante – The disruptor, undercover and seeking emotional truth.

• Melisso – A grounding figure; provides a 'lens of truth' (symbolized through tech).

• Oronte – A bitter gatekeeper or mod figure, angry at being replaced and ignored.

• Oberto – Reimagined as Astolfo’s brother. He enters seeking answers, emotionally raw and outside the system.

Audience Relationship and Framing

• The audience remains outside the curated world—they are voyeurs, followers, or viewers.

• The fourth wall is intact. Characters do not directly address the audience.

• Projections and metrics (likes, views, filters) can suggest the presence of an unseen digital audience.

• In the final scenes, projections fail, filters collapse, and the illusion shatters—ending with raw, unfiltered performance.

Next: from rich detail to slender script

By this point, I had built a rich, organized world to draw from, pinned down the details (the urn, the ring, the winged horse!?!) and mapped out a plot to follow. In the next blog, I’ll dive into the process of translating the arias and then the recits—and how this tremendous pile of information finally became a working script.