Today was about finishing strong and learning from pain. The Great Gatsby interactive story is complete—49 scenes, fully bilingual, ready to play. But the real story isn’t the completed product. It’s what I learned from fixing the same mistakes over and over again.
The Wins: Great Gatsby Goes Live
The numbers look good on paper: 49 AI-generated scenes bringing 1920s Long Island to life. Nick Carraway, Jay Gatsby, Daisy Buchanan—all with consistent faces across hundreds of party scenes and dramatic confrontations. The story works in both English and Korean, with 17 different endings depending on your choices.
But those numbers hide the messy journey underneath.
The Character Costume Disaster
Here’s where things got interesting. After generating the first batch of images, I noticed something weird: Nick and Gatsby were wearing identical outfits in the same scene. Same hat, same necktie, same suit. Like they’d coordinated their wardrobes before the party.
The bug? The AI was trying to maintain “exact costumes” from reference images but couldn’t tell which costume belonged to which character. It saw both characters’ reference photos and just… picked one outfit for everyone.
Three attempts to fix it with better prompts. Three failures.
The solution? Stop asking the impossible. Instead of “maintain exact costume,” I separated character references explicitly: “Character #1 (Nick) - use reference #1 for face and hair ONLY, NOT clothing. Character #2 (Gatsby) - use reference #2 for face and hair ONLY.”
Sometimes fixing a bug means admitting your original approach was fundamentally wrong.
When “One Fixed Costume” Breaks Reality
Then came the bigger design flaw. I’d set up each character with one costume description: “Nick wears simple summer suits with white shirts.” Great for consistency. Terrible for storytelling.
Because here’s the thing: The Great Gatsby happens over an entire summer. Multiple seasons, different contexts—formal parties, intimate dinners, tense confrontations. Real people don’t wear the same outfit for three months straight.
The fix was conceptual, not technical: Stop treating costumes as fixed attributes. Instead, give each character style guidelines: “Always impeccably dressed in expensive tailored suits, favors light colors.” Let the AI adapt the outfit to context while maintaining their fashion sense.
Gatsby can now wear a different suit at his mansion party than at the Plaza Hotel showdown. It’s still obviously Gatsby’s style, but it’s not the same literal outfit.
The Korean Deployment Groundhog Day
But the most frustrating part? Korean deployment issues. Again.
Wrong text tone. Missing HTML structure. Broken image paths. These weren’t new problems—I’d fixed them before in previous story packs. But there was nothing stopping them from happening again. And again. And again.
After fixing the same personality calculator bug for the third time today, something clicked: I was treating symptoms, not the disease.
The disease was lack of prevention systems. I could fix individual bugs all day, but without a checklist and automated validation, I’d just keep fixing the same things forever.
So I built what I should’ve built weeks ago:
- A comprehensive checklist documenting every Korean deployment requirement
- An automated validation script that checks 6 critical issues before deployment
- Updated core scripts so they generate correct output by default
Now before deploying any Korean story, I run python3 validate_korean_deployment.py. If it passes all checks, deploy. If not, fix first. Simple.
The validation takes 10 seconds. It could save hours of debugging.
What I Learned Today
Good code isn’t about never making mistakes. I made plenty today. It’s about learning from them and building systems so you don’t make the same mistake twice.
The costume bug taught me to question my assumptions. The flexible costume system taught me that constraints need to be thoughtful, not arbitrary. The validation script taught me that prevention is so much easier than repeated fixes.
Tomorrow I’ll probably discover new ways things can break. But at least these particular ways are now impossible.
What’s Next
The Great Gatsby proves the complete pipeline works: story generation → AI images with character consistency → bilingual content → deployed website. All the painful lessons are now baked into the system.
Phase 2 starts soon: Regenerating the existing stories (Jekyll & Hyde, Frankenstein, Pride & Prejudice) with these new quality standards. They deserve the same level of polish.
But tonight? Tonight I’m playing through The Great Gatsby to see if Nick’s choices can save Gatsby from his tragic fate. Sometimes the builder gets to enjoy what they built.
What If Classics is an interactive storytelling platform turning classic literature into choice-driven experiences. Each story takes 3-5 minutes to play but branches into dozens of unique endings. Follow the journey at whatifclassics.com.