By Taki Hasegawa
We’ve all seen them: the stock photo of the smiling customer service agent who looks like they’ve never had a bad day. Or the frustrated customer whose expression is so exaggerated it borders on caricature.
Training through roleplays, like those offered through ReflexAI’s Prepare platform, is a powerful and immersive learning experience. But using generic, staged images can break the illusion of reality and dilute the impact of your simulations. The good news is that the generative AI tools we have access to today give us the power to create photorealistic avatars.
This guide will walk you through a detailed process for creating compelling, context-aware AI avatars that will help your team connect with your customer personas on a human level.
Why a face matters: The science behind visual empathy
When your agents are in back-to-back calls, it’s easy for the person on the other end of the line to become an abstraction: a case number, a problem to solve, or a voice in the queue.
Introducing a realistic face interrupts that pattern. In fact, studies show that the human brain is hardwired to respond to faces. Seeing a person’s expression, the subtle signs of stress or fatigue around their eyes, or the environment they’re in, triggers a more profound emotional and empathetic response.
This visual connection can close the psychological distance between the agent and the customer. In other words, the interaction evolves from resolving the ticket for “Case #74B” to helping a real person who is tired, sick, or confused. And this shift is the foundation of true empathy.
The foundation: Building your persona’s story before the prompt
Before writing a single word, it might help to step into the customer’s shoes and consider:
- Who is this person? What is their approximate age, background, and emotional state right now? Are they calm, anxious, frustrated, confident?
- Where are they calling from? Is this an in-person interaction, a video call, or a phone call? Are they in a professional boardroom, a cluttered home office, their car, or a hotel lobby?
- What is their socioeconomic status? The context of a person’s life shapes their reality. An American Express Centurion cardholder calling from their home office likely has a different avatar background than someone calling from their studio apartment.
Think about the details. One might have high ceilings, crown molding, and a Herman Miller chair. The other might have an air conditioner in a window that looks out onto another building’s brick wall.
Going through this exercise can yield richer details to inform your avatar prompt.
The anatomy of a perfect AI avatar prompt
Once you have a clear story, you can translate it into a language the AI can understand. We recommend using a structured format to make sure you don’t miss any critical details. Here’s a template you can use to get started:
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers:
- Image style: photorealistic, 2D illustration, 3D illustration
- Mood: approachable, professional, playful, serious, angry, calm, sad, happy, unhappy
- Framing: close-up, headshot, bust/medium close-up, half body, full body
- Communication Device: Headset, in-ear headphones, cellphone, laptop only, traditional phone
- Angle & vantage: eye-level, slightly above, slightly below, angled upward from lap, over-the-shoulder
- Persona Background: Age, gender, race, hair, facial features, clothing
- Environment: Location, objects, decorations
- Gaze: At the camera, away from the camera
- Use attached photo as background: Yes / no (See the next section!)
- Lighting (optional):
- Lighting sources: window light, desk lamp, ring light, overhead fluorescent, softbox, screen glow
- Lighting temperature: warm (golden), neutral, cool (blue)
- Lighting key: high‑key (light/airy), mid‑key (balanced), low‑key (dark/moody)
- Depth of Field (optional): shallow (subject sharp, background blurred), medium, deep (everything sharp)
Pro-Tip: Finding inspiration for your environment
If you’re struggling to visualize the perfect background for your persona, use websites like Pinterest and Houzz for visual inspiration. For example, if you’re creating a persona in a home office, browsing a collection like this one on Houzz can give you specific ideas for furniture, lighting, and decor.
You can also use a large language model (LLM) to do the descriptive work for you. Find an image you like on one of these sites, upload it to a multimodal AI (like Gemini or ChatGPT), and ask it to describe the scene in detail.
You can then copy and paste that text directly into the ‘Environment’ section of your avatar prompt. It’s important to use the text description instead of uploading the reference photo directly into your final avatar prompt. Directly using an image as a background reference can sometimes cause the AI to create an artificial-looking composite, which can break the sense of realism.
From theory to practice: 5 real-world scenarios & prompts
Let’s put it all together. Here are five distinct scenarios, each with a persona story and a detailed, ready-to-use prompt built from our template.
Scenario 1: The telehealth call
Persona story
Michael, 38, has had a nasty cold for over a week. He feels terrible and is calling a telehealth service from his bed to get antibiotics before a big work meeting.
AI prompt
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers. Please provide a 1:1 aspect ratio image based on the specification below.
- Image style: Photorealistic (webcam look: soft focus, slight digital noise)
- Mood: serious; sad; unhappy; exhausted
- Framing: bust/medium close-up
- Aspect ratio: 16:9
- Orientation: landscape
- Communication Device: laptop only
- Angle & vantage: angled upward from lap
- Persona Background: Age 38; gender male; race Caucasian; hair messy; facial features scruffy beard; clothing faded grey T‑shirt; looks exhausted/sick
- Environment: propped against a simple wooden headboard with white bedsheets visible.
- Gaze: looking slightly downward toward the webcam.
- Use attached photo as background: No
- Lighting sources: desk lamp (off-frame)
- Lighting temperature: warm (golden)
- Lighting key: low‑key (dark/moody) Depth of Field (optional): shallow (subject sharp, headboard slightly out of focus)
- Depth of Field: shallow (subject sharp, headboard slightly out of focus)
Scenario 2: The financial services inquiry
Persona story
Alena, 52, a successful entrepreneur, is calling her private banker from her corner office. She’s confident and focused, wanting to discuss a significant investment in AI stocks.
AI prompt
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers. Please provide a 1:1 aspect ratio image based on the specification below.
- Image style: Photorealistic
- Mood: Professional, approachable
- Framing: Headshot (perfectly centered)
- Communication Device: Laptop only (no headset visible)
- Angle & vantage: Eye-level, from a camera mounted on a large monitor
- Persona Background: Age: 52; Gender: Woman; Race/Ethnicity: Black; Hair: Not specified; Clothing: Tailored navy blue blazer
- Environment: Modern, minimalist high-rise office; Floor-to-ceiling windows with daytime view of city skyscrapers; clean desk with silver lamp and white orchid; Minimalist
- Gaze: At the camera (confident, engaged expression)
- Use attached photo as background: No
- Lighting sources: Window light; even professional key (unspecified)
- Lighting temperature: Neutral (daylight)
- Lighting key: High‑key (bright, even)
- Depth of Field: Shallow (subject sharp; background softly blurred but recognizable)
Scenario 3: The crisis line call
Persona story
Jordan, 19, is calling a crisis line late at night from their dorm room. They feel isolated and overwhelmed. Their face shows deep vulnerability and distress.
AI prompt
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers. Please provide a 1:1 aspect ratio image based on the specification below.
- Image style: Photorealistic; candid, low‑light phone camera look; slight grain
- Mood: Sad; serious
- Framing: Close-up (face)
- Communication Device: Laptop only (off‑camera; not visible)
- Angle & vantage: Eye-level
- Persona Background: Age 19; gender non-binary; race East Asian; hair short, dark; facial features unspecified; clothing oversized dark hoodie
- Environment: Sparse dorm room; plain off‑white wall behind; minimal decor
- Gaze: Away from the camera (looking just past lens)
- Use attached photo as background: No
- Lighting sources: Bright screen glow (laptop)
- Lighting temperature: Cool (blue)
- Lighting key: Low‑key (dark/moody)
- Depth of Field: Shallow (face sharp, background falls out of focus)
Scenario 4: The hospitality check-in
Persona story
David, 45, is at a hotel front desk with his family (out of frame) after a long travel day. There’s an issue with his reservation. He’s not angry, but he is visibly tired and stressed.
AI prompt:
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers. Please provide a 1:1 aspect ratio image based on the specification below.
- Image style: Photorealistic
- Mood: serious, unhappy (tired, slightly frustrated)
- Framing: bust/medium close-up (chest up)
- Communication Device: none (in-person)
- Angle & vantage: eye-level; POV from the front desk agent
- Persona Background: age 45; gender man; race Hispanic; hair unspecified; facial features: tired, slightly frustrated expression; clothing: comfortable, slightly wrinkled travel shirt
- Environment: modern hotel lobby; softly blurred background; objects/decorations unspecified
- Gaze: at the camera (direct eye contact)
- Use attached photo as background: No
- Lighting sources: overhead ceiling fixtures (neutral lobby lighting)
- Lighting temperature: neutral
- Lighting key: high‑key (bright)
- Depth of Field: shallow (subject sharp, background blurred)
Scenario 5: The enterprise software pitch
Persona story
Sarah, 48, an IT Director, is on a video call listening to a sales pitch. She’s professional and attentive but also skeptical, and her expression is neutral and hard to read.
AI prompt:
You are an image generation tool for avatars. The avatars will be used as the profile picture for various call centers. Please provide a 1:1 aspect ratio image based on the specification below.
- Image style: photorealistic (screen‑capture look with slight Zoom compression)
- Mood: professional, serious, calm
- Framing: headshot
- Communication Device: laptop only (webcam on external monitor), no headset
- Angle & vantage: eye‑level
- Persona Background: Age 48; gender woman; race unspecified; hair not specified; facial features: professional glasses; clothing: simple dark red blouse; expression: neutral, attentive, slightly skeptical
- Environment: organized home office; white bookshelf with technical books and a single green plant; background softly blurred
- Gaze: at the camera
- Use attached photo as background: no
- Lighting sources: screen glow (monitor key), window light (side fill)
- Lighting temperature: neutral
- Lighting key: mid‑key (balanced)
- Depth of Field: shallow (subject sharp, background blurred)
From generic AI avatars to human connections
Creating a truly realistic AI avatar isn’t about mastering the AI; it’s about deepening your understanding of the people you serve. By moving beyond generic prompts and embracing the specific, nuanced details of a person’s life and context, you create more than just a picture. You create a powerful training tool that equips your team to build genuine, human-to-human connections.
At ReflexAI, we believe that realistic practice is the key to building mastery. Integrating these kinds of detailed, empathetic personas into your training scenarios is a critical step toward preparing your team for the real world.
