Roleplay Sessions allows you to add practice, coaching, and measurement to your Synthesia projects to create one end-to-end learning loop. Learners step into a real conversation with an Avatar that talks, asks questions, and pushes back, then get coached and scored in real time, with their skill uplift measured over time.

Ways to create a Roleplay Session

There are three ways to get started: use Assistant, start from a template, or add a Roleplay scene to a video.

1. Use Assistant to create a Roleplay Session

   From the Home page, select Assistant, choose your roleplay type, and describe your scenario in a prompt. Add one or more files as context if desired. After Assistant builds the roleplay for you, you can customize and iterate on it from the editor, and continue collaborating with Assistant via chat.
   
   Learn more about creating a Roleplay Session with Assistant.

   

2. Start from an agentic template

   Select an agentic template from the template picker and customize it from there.

   

3. Add a Roleplay scene to your video

   Add a Roleplay scene to any existing or new video directly in the editor to get started. You'll also need to add and configure a Feedback scene. You can continue to edit and configure your content manually or collaborate with Assistant via chat. Learn more about scene types.

Scene types

Video, Roleplay, and Feedback scenes are the building blocks of Roleplay Sessions.

To add a scene to your Session:

1. At the top of the scene list on the left, select + Add scene.
2. In the template panel, choose one of the three scene types: Video, AI Roleplay, or Feedback.
3. Select a scene type to add it to your project.

Video scene

A Video scene is the standard Synthesia scene type: write the script, choose your avatar, and it plays straight through. In a Roleplay Session, Video scenes are typically used to brief learners and explain what they're about to practice before they enter the Roleplay scene.

Roleplay scene

A Roleplay scene is where the learner practices. It's a live, real-time conversation: your learner speaks, and the AI persona responds, pushes back, and reacts like a real person would.

There's no script for this scene type. Instead, you configure:

• Who the AI persona is
• What situation the learner is walking into
• The skills you're measuring

See Roleplay scene configuration to learn more.

Feedback scene

Feedback scene comes after the Roleplay scene. Synthesia automatically generates a score and personalized coaching based on the passing criteria and skills you set in the Roleplay scene, so learners can see what went well and what to improve.

The Feedback scene isn't static: learners can talk to the coach. They can ask follow-up questions, dig into specific moments in the conversation, and get further guidance.

See Feedback scene configuration to learn more.

Roleplay and Feedback scene configuration

This section walks through every component of the Roleplay and Feedback scenes. When you start from a template or Assistant, much of the setup is already done for you, but it's important to review (and revise, if necessary) the Roleplay and Feedback scene configurations in order to ensure you're driving the results you want to see from your learners.

Roleplay scene configuration

There are six components to set up when configuring a Roleplay scene:

Roleplay type

Roleplay type sets the context for the roleplay and tells the AI what kind of conversation it's running.

Click the Roleplay type dropdown at the top-left of the script box to select an option:

Choose the Roleplay type that best matches your scenario from one of the following available options:

AI Agent persona

AI Agent persona is who your learner will be speaking to during the roleplay.

Choose from a library of pre-built personas or build your own from scratch. You can customize the details for any of the pre-built personas available in the persona library to fit your scenario, so feel free to use it as a starting point rather than starting from scratch.

Basics

Start with the persona's name, job title, and company. These ground the persona in a believable context.

Base personality

Set the persona's default tone for the conversation, ranging from skeptical and blunt to curious and empathetic. Choose based on who your learner is actually facing in the real world. The more closely this mirrors reality, the more useful the practice.

Available personalities:

Background

Background is where you define the persona's stable traits—things that hold true across the whole conversation, such as personality, communication style, or general attitude. For example, in a customer service scenario you might describe the persona as a long-time customer with a direct communication style who values efficiency.

Scenario instructions

Scenario instructions set the specific situation the learner is walking into. Write these as instructions to the AI persona, telling it how to handle the conversation (they are not instructions for the learner themselves).

The more precisely you write these instructions, the more authentically the AI responds, and the more realistic the practice feels to the learner.

Strong instructions are specific and cover:

• What's happening in the scenario
• The persona's motivations
• How the conversation should unfold
• How the persona should interact with the learner

Objections

Objections introduce challenge into the conversation. They're specific questions the AI persona will raise and pushback the learner will need to handle.

Adding objections:

• Add one objection per line. Each line you type becomes a separate objection (press Enter on your keyboard to start the next one).
• Base these on the real points of pushback that come up in the conversations your learners are preparing for. They should also be realistic for the specific persona you're working with and line up with the background and personality you've already defined.
• 3-6 objections are plenty for a single Session.

Skills

Skills are how your learner's performance is scored. Each skill you add to a Roleplay scene is evaluated by the AI once the conversation ends. Skills should be observable behaviors rather than abstract qualities. Stick to specific, visible actions that can be evaluated.

Add three to five skills that align with the specific scenario you've designed. If you're training a customer service rep to de-escalate, your skills should reflect de-escalation, not general communication.

Select Add in the Skills section to open the skill picker. You have three options:

Creating a new skill

Select + New skill to create a new skilll. You'll set it up in order:

1. Set up your scoring levels: before anything else, confirm the scale you'll score this skill against. You'll start with 4 default levels: Not demonstrated, Below expectations, Meets expectations, Exceeds expectations. Scoring levels are the bands the AI grades a skill against. Levels are set per skill, so different skills in your Workspace can use different scales if needed.

   

   1. Add more levels if you need finer-grained scoring (up to 6 score levels max)
   2. Edit the labels for the score levels if needed
   3. Once you're satisfied, click Confirm

2. Name the skill with a clear, specific title (e.g. Active Listening or Objection Handling)

3. Add a description (one or two sentences summarizing what the skill covers)

4. Define performance at each level: describe what that level of performance actually looks like. This is what the AI matches the learner's performance against to determine their score.

5. Optional: Select Configure levels at any point to go back and adjust the number of levels or their labels before saving.

Editing a skill

Select the pencil icon on any skill to edit it. What happens next depends on where the skill came from:

• Stock (Synthesia) skills: you can't edit the original directly.

  Select Copy skill first to create your own copy. You're then free to edit the name, description, and scoring levels. The stock skill itself is never changed.

  

• Workspace skills: these were already saved to your library, so you can edit them right away.

  Select Edit skill to make adjustments. Editing a Workspace skill only lets you adjust the performance descriptions at each scoring level. You can't rename the skill, change its description, or reconfigure the scoring levels themselves from here.

  

Once you've made your changes, choose whether to save them just for this Roleplay scene, or update the skill in your Workspace library so the change carries over everywhere else it's used.

Scoring

Each skill in a scene has a weight (%) that controls how much it counts toward the overall score. By default, skills are scored and weighted equally. Adjust any skill's weight to reflect what matters most and the others will rebalance automatically to 100. Select the lock icon on a skill to hold its weight steady. Any skill left unlocked will auto-adjust so the total always comes out to 100%.

Removing skills

Select the minus icon on a skill to remove it from the scene.

Time limit

Use the Time limit setting to match how long the conversation would realistically last. For example, if a customer support call typically runs five to seven minutes, set the limit there. This keeps the practice realistic and engaging. Whatever you set, be sure to communicate the time limit to your learners in advance in a Video scene so they know what to expect.

Learn more about configuring Roleplay scenes in our Knowledge Base.

Feedback scene configuration

The Feedback scene helps learners reflect on their performance and see a path to improvement.

Select Feedback from the Add scene menu, and Synthesia handles the scoring details automatically based on the skills and rubric defined in your Roleplay scene. You only need to decide two things:

• Pass score
• Whether to allow retries

Pass score

Set the pass score in accordance with your goals for the Session. Generally, we recommend pass scores of 65-70% for practice scenarios and 75-85% for more formal assessments.

Retries

Retries allow your learn to practice and apply their feedback. Whether to allow them depends on your goals for the Session.

Preview, generate, and publish your Roleplay Session

Once your Roleplay Session is configured, three steps remain: preview it, generate it, and publish it to get it in front of your learners.

Before generating, select the Play button in the top-right of the toolbar to preview the Roleplay Session. Run through it as a learner would and check that:

• The persona is behaving as intended
• Objections are landing naturally
• The overall flow feels right

If anything's off, adjust your Instructions or Objections first, as these have the biggest impact on how the conversation unfolds. When you're satisfied with the experience, select Generate. Synthesia checks your script and scenes for configuration errors and gives you a chance to fix any that come up. If no issues are found, processing starts right away.

Once generated, you have four ways to share your Session:

1. Publish (recommended): Send the Session to the Synthesia Learner Portal, a dedicated space for assigning and tracking Roleplay Sessions at scale.
   
   Learn more about publishing Sessions and managing published Sessions.
2. Share page link: A direct link to interact with the full Session.
3. HTML embed: Add the Session to a webpage or your intranet.
4. SCORM package: Upload to your LMS for completion tracking.

Once a Session is published to the Learner Portal, Enablement or Organization admins can assign it to individual learners or groups, track completion and scores, and manage learners all in one place.

What's next

Learn more about the Learner Portal, the Enablement Admin role, and assigning and tracking Sessions.