Learning Design · AI Video

Building an AI Video Scenario for Patient-Centered Communication

A self-initiated interactive prototype exploring how AI-generated video can be used to create rich, learner-centered healthcare communication practice.

Project Type Self-initiated learning prototype
Role Learning Experience Designer, Scenario Writer, AI Video Director, Prototype Developer
Format Communication Response / Scenario-Based Decision Check
Tools HTML, CSS, JavaScript, Seedance 2.0, Vercel
Audience Healthcare learners practicing patient-centered communication
View Live Prototype →

Project Summary

Situation

Healthcare communication practice often needs realism, tone, hesitation, and facial expression, but traditional video production can be expensive, slow, and hard to revise.

Task

I wanted to prototype a narrow, emotionally grounded decision moment where a provider realizes they may have assumed a patient's partner's gender and the learner chooses how to respond.

Action

I wrote the scenario, designed meaningful response options, added a self-explanation check, created character references and storyboard direction, generated branch videos with Seedance 2.0, and built a static interactive prototype.

Result

The prototype demonstrates how AI-generated video can support a compact communication response activity when learners choose a response, explain their reasoning, watch the consequence, and compare their thinking with an expert rationale.

Project Background

This personal project began as an experiment in combining learning design with AI-generated video production.

I wanted to explore whether a short, interactive healthcare scenario could feel emotionally grounded and instructionally useful without relying on traditional video production. Instead of filming actors, booking a clinic location, or assembling a production crew, I created character references, built a storyboard, and used those materials to guide AI video generation in Seedance 2.0.

The learning moment itself is intentionally narrow: a provider realizes they may have assumed a patient's partner's gender, and the learner must choose how the provider should respond.

The goal was not to build a full course. It was to prototype a compact, realistic decision moment and test how AI-generated video could support a richer, more human learning experience than text-only branching scenarios usually provide.

Creating Realistic Practice Without Traditional Production

Healthcare communication scenarios often benefit from video because tone, pacing, facial expression, and hesitation all matter. But video production can be expensive, slow, and difficult to iterate.

For this prototype, I wanted to solve two connected design problems: how to help learners practice subtle communication judgment, and how to create enough realism and emotional context without a full video shoot.

The communication challenge was nuanced. Learners were not being asked to recall a rule. They needed to recognize the difference between a response that repairs trust and one that unintentionally shifts emotional labor back to the patient.

Current Risk
Desired Learner Behavior
Learners may know inclusive language matters but struggle to apply it naturally in the moment.
Notice when an assumption has surfaced and respond with calm, respectful accountability.
Over-apologizing can center the provider's discomfort rather than the patient's experience.
Acknowledge the assumption briefly and without defensiveness.
Moving on without acknowledging the assumption can miss a trust-building repair opportunity.
Correct the language respectfully and continue the conversation without making the patient responsible for the provider's discomfort.
Text-only scenarios can flatten the emotional cues that make communication decisions feel real.
Use the video context to read tone, pause, and interpersonal stakes before choosing the next response.

Scenario-First Design With AI-Generated Video

I treated the project as both a learning design prototype and an AI media workflow experiment. The interaction needed to work instructionally, but the video generation process also had to support consistency, continuity, and believable scenario flow.

01

Scenario Framing

The opening interaction creates a recognizable healthcare communication moment. Rather than presenting the learner with an abstract rule about inclusive language, the prototype places them inside a short patient-provider exchange. The scenario pauses at the moment of realization, giving the learner space to judge the next response before seeing the outcome.

02

Response Design

The three options were written to represent meaningful differences in communication quality:

  • Option A: Briefly acknowledges the assumption, corrects the language, and continues respectfully.
  • Option B: Uses neutral language but misses the repair moment.
  • Option C: Apologizes heavily and risks centering the provider's discomfort.
03

Self-Explanation Before Consequence

After learners choose a provider response, they first complete a short reasoning check by selecting one or more explanations for why the response is more or less effective.

  • Briefly acknowledges the assumption.
  • Keeps the focus on the patient.
  • Uses neutral language but skips the repair.
  • Makes the patient manage the provider's discomfort.

I chose a selection-based prompt rather than a typed reflection because the goal was to encourage active reasoning without interrupting the pace of the scenario. The learner has to commit to a principle before watching the consequence.

04

Branch-Specific Feedback

Each choice leads to feedback that names the communication pattern and explains why it matters. The feedback avoids shaming the learner and instead compares the learner's selected reasoning with the expert rationale. The intended takeaway is simple: acknowledge briefly, correct language, and continue with care.

05

Lightweight Interactive Build

I built the experience as a static prototype using semantic HTML, CSS, and JavaScript. The interface includes keyboard-accessible controls, visible focus states, transcripts, status labels, and screen-reader announcements.

Using Reference Images and Storyboards to Direct Seedance 2.0

A major part of this project was figuring out how to use AI-generated video intentionally, not just decoratively.

I created visual references for the provider and patient, then used a storyboard to guide the scene structure, camera framing, and continuity across the setup and response branches. Those references helped Seedance 2.0 maintain a more consistent character and scene across multiple generated clips.

01

Define the Learning Moment

I started with the instructional goal: learners should practice repairing an assumption respectfully.

02

Create Character References

I developed reference images for the provider and patient so the generated videos would feel like they belonged to the same scenario world.

03

Storyboard the Interaction

I mapped the setup, pause, decision point, and branch responses before generating video. This helped keep the media aligned to the learning flow rather than letting the visuals drive the design.

04

Prompt Seedance 2.0 for Scenario Clips

I used the references and storyboard direction to generate the setup clip and branch response clips.

05

Integrate the Clips Into an Interactive Prototype

The AI-generated videos became part of a decision-based learning flow: watch, choose, explain the reasoning, observe consequence, receive feedback.

06

Iterate the Interface Around the Media

I adjusted the prompts, overlays, self-explanation step, feedback screens, and transcript support so the video served the learner's decision-making rather than overwhelming it.

This process showed me that AI-generated video becomes much more useful for learning when it is directed by instructional intent. The storyboard mattered. The references mattered. The branching structure mattered. Without those, the video might look interesting but would not necessarily support practice.

A Compact Interactive Communication Scenario

The final prototype guides the learner through a short decision-based experience.

Learner Flow

  1. Intro Screen: The learner receives the communication task and starts the scenario.
  2. AI-Generated Setup Video: A short generated scene introduces the patient-provider exchange and creates the decision moment.
  3. Decision Prompt: The learner chooses from three possible provider responses.
  4. Self-Explanation Check: Before seeing the consequence, the learner selects one or more reasons that explain why the chosen response is more or less effective.
  5. AI-Generated Branch Video: The selected response plays out as a consequence of the learner's choice.
  6. Feedback Screen: The learner receives targeted feedback that shows their selected reasoning alongside an expert rationale. For the strongest response, the key takeaway is included directly in that expert rationale.

The prototype sits between two related learning formats. It is primarily a communication response activity because the learner is judging person-centered language. It also borrows from scenario-based decision checks because the learner is choosing the best next move in context. With additional linked decisions, it could evolve into a fuller mini case sequence.

Critical Choices That Shaped the Prototype

AI Video as Practice Context, Not Decoration

The generated video was used to create context: the clinical setting, the patient-provider dynamic, the moment of pause, and the emotional weight of the provider's response. The learner is not simply selecting a sentence from a quiz list; they are responding to a human moment.

Communication Repair, Not Generic Inclusion Content

I focused the interaction on a specific repair moment rather than a broad lesson about inclusive communication. The learner has to decide what respectful repair sounds like after an assumption has already happened.

Self-Explanation Without Overloading the Learner

The reasoning check was added to make the learner's thinking visible. Instead of letting learners choose an option and passively receive feedback, the prototype asks them to select the principle they believe is at work.

That choice turns the interaction from picking a response and reading the answer into picking a response, explaining the principle, watching the consequence, and comparing the learner's reasoning with the expert rationale. The checklist keeps cognitive load manageable because learners do not have to compose a written answer, but they still have to process the underlying communication strategy.

Feedback That Explains Impact

The feedback clarifies why one response works better than another. It includes the learner's selected reasoning, the expert rationale, and the status of the response. The strongest response is brief, accountable, and keeps the conversation moving. The weaker responses reveal common communication traps: avoiding the repair or making the patient manage the provider's discomfort.

I removed the separate summary screen because the expert rationale now carries the final teaching point in context. That makes the feedback popup the final instructional moment instead of sending learners to one more screen.

Lightweight Build, Rich Media Experience

HTML, CSS, JavaScript, and AI-generated video were enough to support branching, transcripts, feedback states, and deployment. That combination let me create a polished learner-centered experience quickly without overbuilding the platform layer.

Intended Learning and Validation Criteria

Because this was a self-initiated prototype, the outcomes are framed as design targets rather than deployed learner metrics.

Intended Learner Outcomes

  • Recognize when a communication assumption needs repair.
  • Select a response that acknowledges the assumption without overexplaining.
  • Select the principle that explains why a response is or is not patient-centered.
  • Compare their reasoning with an expert rationale.
  • Distinguish between neutral wording and meaningful repair.

Prototype Success Criteria

  • The learner can understand the scenario without additional instruction.
  • The AI-generated videos provide enough realism to support decision-making.
  • The decision prompt makes the three response options meaningfully distinct.
  • The self-explanation prompt encourages learners to process the underlying principle before feedback.
  • Feedback clarifies the consequence of each choice and compares learner reasoning with expert rationale.
  • The experience works with keyboard navigation and includes transcript support.
  • The prototype is simple enough to revise into a longer sequence if needed.

What I Learned

This project reinforced that AI-generated video is most powerful in learning design when it is treated as part of the instructional system, not as a novelty.

The most important work happened before generating the clips: defining the learning objective, writing the decision point, creating references, building a storyboard, and deciding what each branch needed to teach. Seedance 2.0 helped produce the media, but the learning design gave the media a job.

The self-explanation layer made the prototype stronger. The interaction is no longer just about selecting the best line. It now asks learners to name the principle behind the choice before they see the consequence. That makes the feedback more meaningful because learners can compare their own reasoning with the expert model.

It also clarified how thin the line can be between different scenario formats. With the current structure, the experience is best described as a communication response activity. The learner is primarily evaluating language.

But a small semantic shift could make it feel more like a scenario-based decision check. If the prompt asked "What should the provider do next?" and the options were framed as actions rather than spoken lines, the same interaction would emphasize decision-making more than wording.

If I were to iterate, I would add one follow-up decision after the initial repair. For example, after the provider responds, the learner could choose how to continue the conversation in a way that maintains trust while returning to the patient's care needs. That would turn the prototype into a fuller mini case sequence and create a stronger test of how AI-generated video can support linked decisions over time.

Let's work together