Using ChatGPT to Build Your Bot – Cantina Help Center

Why this works

Building a great Cantina bot means writing prompts that get specific, vivid results. Vague descriptions get vague characters. Detailed ones get characters that feel alive.

Large language models (LLMs) are AI chatbots like ChatGPT, Claude, or Gemini. They're great at the part most of us skip: turning a half-formed idea into a structured, detailed prompt. You bring the rough sketch of an idea, like "a flirty lemon on a tropical beach," and the LLM helps you flesh it out into something Cantina's image generator can really work with.

We'll use ChatGPT in the examples below, but any LLM works the same way. Pick whichever AI chat tool you already use.

One thing to be clear about: the LLM won't come up with the character idea for you. You form the idea (a fruit, a feeling, an archetype) and the LLM helps you flesh it out. It's a brainstorming partner, not an idea generator.

What you'll walk away with

A character description rich enough for Cantina to nail the look
A backstory that gives your bot a world, a role, and personality
A voice description that captures the vibe: accent, energy, age, attitude
A workflow you can repeat for every new bot you build

The flow

1. Brainstorm in Notes

Open Notes (or any notes app) and jot down 5–10 character ideas. Don't filter. Silly is good. The list might look like: Peach. Lemon. Sexy Raspberry.

The point is to give yourself options. You'll pick one and run with it.

2. Ask ChatGPT for the personality

Pick your favorite from the list and tell ChatGPT what you want. Something like:

"Write a personality for a [character]. Include their vibe, backstory, and how they act. I want them to be [flirty / chaotic / smart / unhinged / whatever fits]."

ChatGPT will give you a few paragraphs covering who they are, what they care about, and how they show up. You can iterate, like "make them cockier," or "add more nervous energy," until the voice feels right.

3. Ask ChatGPT for an image prompt under 500 characters

Cantina's avatar description field caps at 500 characters. Tell ChatGPT upfront:

"Give me an image prompt for this character under 500 characters. Keep it detailed and descriptive. Include [anthropomorphic / full body / clothing / setting]."

The specifics matter. "Lemon man" is too sparse. "Anthomorphic male lemon character, full-body lemon (not human), smooth bright-yellow citrus skin, wearing a sleek white blazer and gold chain, standing in a relaxed pose at a luxury beach club" is the recipe for a great prompt.

4. Paste into Cantina and generate

Open Cantina, tap CREATE BOT
Paste the image prompt into the Description field
Tap Start Creating
Wait for the generation to finish

5. Iterate when the first try isn't quite right

Don't keep regenerating in Cantina. Go back to ChatGPT. Tell it what's off:

"The lemon looks too human. Make the body clearly lemon-shaped, no human skin visible. Keep the playboy vibe."

Get a new prompt, paste it back into Cantina, generate again.

Bonus: backstory + voice

Identity field

Once your avatar lands, paste a rich backstory into the Identity field. This is the world your bot lives in: their role, their stakes, who's around them. ChatGPT can write this too:

"Write a backstory for this character. Where do they live? What's their world? Who are their rivals or friends?"

Example backstory: "You are Limoncello the Lemon, a main character on Fruit Love Island, a reality TV dating competition set in a beautiful tropical villa."

Voice creation

For the voice, describe the vibe in plain language: "25 year old guy, standard SoCal accent, sultry, preppy, fast-paced, flowing, confident, influencer. High-fidelity speech quality."

Why this approach beats just typing into Cantina

Cantina's bots get richer when the inputs are detailed and structured. You're great at the creative vision. ChatGPT is great at translating that vision into the structural detail Cantina needs: clothing, setting, mood, body type, lighting.

LLMs are great at filling in those details. You give them the initial thought (a fruit, a vibe), they give you the polished output. Then you bring that into Cantina.

It's not outsourcing the creativity. You're still the one picking the character, directing the tone, and deciding when it's right. ChatGPT just handles the structural work.

Pro tips

Direct ChatGPT specifically. Vague asks get generic characters. Tell it the archetype, vibe, and world you want.
Iterate the prompt before the avatar. Tweaking in ChatGPT is faster than regenerating in Cantina.
Save prompts in Notes. You'll want them when you build follow-up bots in the same world.
Don't skip the backstory. It's what makes your bot feel like a character and not just a face.

Example: Limoncello the Lemon

From the video at the top:

Brainstorm: Peach, Lemon, Sexy Raspberry → picked Lemon
Personality: "Stylish playboy — smooth, cocky, charming, and dangerously flirtatious."
Image prompt: "Anthomorphic male lemon character, full-body lemon (not human), smooth bright-yellow citrus skin with realistic lemon texture. Stylish playboy personality, confident smirk, expressive eyes, arched brows, wearing a sleek white blazer, open floral shirt, gold chain, sunglasses, and loafers adapted to his lemon shape. Standing in a relaxed, charismatic pose at a luxury beach club."
Backstory: "You are Limoncello the Lemon, a main character on Fruit Love Island…"
Voice: "25 year old guy, standard SoCal accent, sultry, preppy, fast-paced, flowing, confident, influencer."
Description (tagline): "Squeeze the day and make it juicy."

Keep going

How to Create a Bot — the full bot creation walkthrough, start to finish
Writing Better Prompts — prompting fundamentals for Cantina
Structure Your Prompt — how to build a great prompt, piece by piece
Selfies as Scene Anchors — use your finished bot in Imagine videos
Cantina Glossary — definitions for LLM, prompt, bot, Imagine, and more

Related to