Prompting 101

Places you'll use prompts

You'll run into prompts in a few places:

• Bot creation — fields like identity, personality, and backstory are prompts themselves. Write them like you're describing the bot to the model. If you use the Magic Bot Creator instead, your short inputs are turned into those prompts for you, and you can still open and edit them after.
• Bot voice — describe the voice you want (gravelly narrator, breathy whisper, perky game show host) and the app uses that prompt to generate it.
• Selfie prompts and drawings — the free-form text you type when you ask a bot to take a selfie of itself, or to draw something else for you.
• Imagine — the video tool. Imagine takes a prompt and generates whole new images and a video around it.

Prompting your image or selfie

The model reads your words literally and paints what you describe. The more of the scene you give it, the more your result will look like what you pictured.

• "Selfie tacos." With just two words, the model picks the scene for you. Are you eating tacos? Cooking them? Wearing one as a hat? Anything goes.

• "Selfie of me eating a taco at a beach taco truck at sunset." Now there's a setting, an action, a time of day, a vibe. Way closer to what you wanted.

  

Rule of thumb: the more your prompt sounds like the photo, video, or reply you're picturing, the closer the model can get.

Three quick wins

1. Use the shadow text as inspiration. The grayed-out text inside an input box is a hint showing what to write — sometimes a full example labeled "E.g.", sometimes a shorter cue like "Selfie of me…" or "Imagine…". It disappears when you tap in and start typing. Write your own prompt using it as a guide. 

   

2. Open the prompt and edit it. Anytime the app generates a prompt for you (that's Cantina Prompt Magic working in the background), you can tap to expand and edit the actual text the model will see. Those generated prompts are tuned for good results, so reading them is one of the fastest ways to pick up what a great prompt looks like. Then edit anything you want to change.

   

3. Lean into specifics. Concrete details give the model more to picture from. "Pink-and-orange sunset over the ocean, shot on an iPhone" gets you closer to the image in your head than "beautiful sunset photo."

When to use chat vs Imagine

• Short, in-the-moment request — a quick line is enough. The bot can riff off your message and reply fast.
• Brand-new scene with brand-new visuals — go to Imagine. It takes longer but generates everything from scratch.

Keep going

• Fast Videos vs Imagine Videos — when to use each
• Cantina Glossary — the words and product names you'll see across Cantina
• Bot Selfies & Themes — making bot-of-itself images
• Video Editor 101 — building a video scene by scene