OLGA IN THE CORNER
An AI Podcast Journey (Part 2)
In the previous instalment, we introduced the concept and goals of developing a viable, scalable AI production pipeline for animation and video. In this article, we cover the preproduction phase, focusing on:
- Story
- Character Design
- Voice Over (V.O.)
- Rough Edit
We’ll look at where AI really helped lighten the workload—and where it just didn’t cut it. (Not “useless” as in broken—just not particularly useful for the specific task at hand.)
In the final instalment, we’ll dig into animation, editing, branding, and publishing the first episode.
1. STORY
I initially tried to salvage a script that ChatGPT generated, but it ended up too scattered to be useful—and I had no clear idea how long the episode would run once all the action was packed in.
To keep things within our 2-minute format, I rewrote the script as pure dialogue: two A4 pages, double spaced. No descriptions, just character lines.
👉 Notebook LM
Before I begin writing, I like to get a different perspective on the concept. For that, I used Google’s Notebook LM—a seriously underrated tool.
Notebook LM can import PDFs, URLs, YouTube videos, and more—then summarise them from any point of view you request. One feature I really like is its podcast-style discussion tool, where two AI characters “talk” about your topic.
The output helped spark new ideas, highlighted flaws, and surfaced themes I might have missed. You can resubmit the same material and it’ll generate a totally different take. There’s even an interactive mode where you can ask questions or add information, and the AI voices respond in real-time. (It’s honestly a little creepy.)
👉 Script
I considered writing the script in proper industry format using Final Draft, but since we were only using it for voice generation, I stuck with basic dialogue. The AI voice generator only needs the lines—everything else would’ve been deleted anyway.
Here’s a link to the dialogue script. As you’ll see, it’s sort of industry standard: courier font, 12pt, double spaced, etc.
2. CHARACTER DESIGN
There are tons of AI image generators out there—some surprisingly good and free—including Microsoft Designer, ChatGPT, Leonardo, and Ideogram.
ChatGPT created our first version of the characters, Dex and Edna. (It mistakenly named Edna as Olga, but hey—easy fix.) With most generators, you describe what you want, and the AI offers several options. You can pick one or continue tweaking until you’re happy.
We found Leonardo especially good at producing rich detail. But we went with Ideogram in the end because it gave us characters with subtle touches we loved—like the natural-looking glare off Edna’s glasses.
3. VOICE OVER
For this project, we used 11 Labs—our go-to for AI voice. It’s reliable and delivers decent results.
One important thing we learned: you have to add space between each character’s dialogue. Otherwise, the AI just ploughs from one line into the next, making the delivery sound weird and unnatural.
As far as I know, 11 Labs doesn’t yet distinguish between different characters in the same script, so each line has to be separated and processed carefully.
What is amazing is their voice manipulation tech:
- Speech-to-text
- Text-to-sound
- Voice cloning
- Real-time dubbing
- Modifying recorded human voices
That last one is a game-changer. Instead of trying to wring emotion from a text-to-voice tool, you can just record a human actor delivering the line with proper performance—then use AI to change the voice’s gender, tone, or accent.
4. ROUGH EDIT / ASSEMBLY
We chose to do the edit manually in Adobe Premiere.
While Adobe has been loading up with AI features lately, the one we found most useful was its auto-captioning tool. It’s not perfect—accents can throw it off—but it saves a ton of time. If you’re curious, there are loads of great YouTube tutorials on how to create stylized captions.
We followed a tutorial on how to create Hormozi-style captions—and it worked great.
There are also dedicated AI editing platforms like AIVideo.com, Filmora, Opus Clip, and Firecut. But as longtime Premiere users (and proud control freaks), we opted to go manual to retain full creative control.
CONCLUSION: AI’s Report Card (So Far)
As we explore different AI-based production pipelines, the goal is to understand where the time and creative savings are—and where the tech still needs training wheels.
Here’s our take on each stage:
🟢 1. STORY
AI was decent at idea generation and showing different perspectives. But subtle humour, cultural insight, and tight structure still needed a human touch.
Rating: ⭐⭐⭐☆☆
🟢 2. CHARACTER DESIGN
AI absolutely crushed it. Realistic, creative, distinctive character designs with minimal prompting.
Rating: ⭐⭐⭐⭐⭐
🟡 3. VOICE OVER
It’s a mixed bag. AI voices still fall short of human emotion, but the ability to clone or modify human recordings makes up for it.
Rating: ⭐⭐⭐½
🟡 4. EDITING / ASSEMBLY
While AI editors are catching up, we stuck with Premiere. That said, Adobe’s captioning tool is a legit timesaver.
Rating: ⭐⭐⭐½
Next up: we dive into animation, branding, and publishing the first episode of Olga In The Corner.
Stay tuned.
