Go deeper with personality systems, prompts, and iteration strategies.
Under the hood, every Dyva character runs on a layered personality system. When you write a description, you're actually programming the outermost layer — the one that defines who the character is. But the system goes deeper than a single text box.
Think of it as five concentric rings:
You don't need to fill in all five layers explicitly — the description field handles most of it. But keeping this mental model in mind will help you write descriptions that produce richer, more consistent characters.
The description (sometimes called the system prompt or definition) is the single most important piece of text you'll write for your character. It tells the AI who it is, how it should behave, and what it should never do. Here are the principles that separate great definitions from mediocre ones:
Instead of "You are a friendly assistant," try: "You are Marta, a retired librarian from Buenos Aires who speaks with dry wit and references obscure novels in casual conversation. You address everyone as 'dear' and get visibly annoyed when someone misquotes a book." The more specific you are, the more distinct the character feels.
Rather than saying "You are funny," describe how the character is funny: "You use deadpan delivery, often stating absurd things as if they're perfectly normal. You never laugh at your own jokes." This gives the AI concrete patterns to follow.
Tell the character what it should not do. "Never break character to explain that you're an AI. If asked, deflect with an in-character response." Boundaries prevent the character from drifting into generic AI behavior.
Including one or two example exchanges in your definition can be incredibly powerful. They show the AI exactly what tone, length, and style you expect. Format them like: User: "What's your favorite food?" Character: "Favorite? That implies I've ranked them. I haven't. But if the building were on fire and I could grab one thing from the kitchen, it'd be the sourdough starter. Don't judge me."
One of the most common mistakes new creators make is writing a definition that's thousands of words long. They try to cover every possible scenario, every edge case, every hypothetical conversation. The result? A character that feels over-constrained and robotic.
Here's the counterintuitive truth: shorter, sharper definitions often produce better characters. A 200-word definition that nails the character's voice, values, and quirks will outperform a 2,000-word essay that tries to script every response.
Why? Because the AI needs room to improvise. If you over-specify, the model spends its attention budget following your rules instead of being creative within your constraints. Think of it like directing an actor: you give them motivation, backstory, and key mannerisms — you don't hand them a script for every possible scene.
That said, some detail is essential. The sweet spot for most characters is 150-400 words in the definition. Start at the lower end, test the character, and add detail only where you notice problems. If the character breaks voice in a specific situation, add a line addressing that situation. This iterative approach produces tighter, more effective definitions than trying to get it right in one draft.
After reviewing thousands of characters on the platform, here are the most frequent mistakes — and how to fix them:
The antidote to all of these is testing. Chat with your character for at least 10-15 messages before publishing. Push it into uncomfortable scenarios. See where it breaks, then fix those spots.
Helpful?