E-Learning Heroes Challenge #557: Consistent AI Characters
For E-Learning Heroes Challenge #557, the task was to share a go-to custom AI image prompt. I chose to tackle the hardest problem in AI imagery for our field: character consistency. Scenario-based e-learning depends on it. When a learner follows one character through a branching scenario, that character has to look like the same person in every scene, at their desk, in a meeting, out in the world. AI models drift, so most generated characters quietly become a different person from slide to slide, which breaks immersion and credibility. Here is how I build a single character and hold them steady across an entire course.
MEET THE CHARACTERÂ
Director of Career Services, Oakridge State University
She is the Director of Career Services, Oakridge State University. Married to Michael, mother to Taylor (7) and Imani (6). Dr. Nichols is an original character I created as an authoritative yet approachable higher education leader with almost 20 years spanning academic advising, student retention, and workforce alignment. She is the bridge between university talent and evolving industry needs, grounded by a full family life. I built her to be usable across both professional and personal scenes, because a good scenario character has to live in more than one setting.
Prompt 1: Build the character and lock the style
Create a premium Pixar-quality 3D corporate illustration with cinematic lighting, sharp focus, ultra-clean rendering, realistic materials, and professional presentation artwork. Rich natural colors, no grain, no blur, no painterly effects, high-end CGI quality, uncluttered composition, executive-presentation ready.
*Consistency note: This first prompt does the heavy lifting. It establishes the character and a precise, repeatable visual style. I generated and iterated in Google Gemini until the character and style were exactly right. Once locked, this becomes the foundation every later image references.
Prompt 2: Change the scene, keep the character
Create an image of Shantell talking to a colleague. Both are seated at a table in a conference room.
*Consistency note: With the style locked in Gemini, I changed only the setting and action, not the character. This is the core technique: hold the person constant, vary the scene around them.
Prompt 3: Prove it holds outside the office
Create an image of Shantell grocery shopping.
*Consistency note: Moving the same character into an everyday, non-work setting is the real test. If she still reads as the same person here, the consistency is genuine, not just a lucky match between two similar office scenes.
Prompt 4: Extend to a group scene
Create an image of Shantell with her family.
*Consistency note: The hardest case is keeping your character consistent while adding other people. Holding Shantell steady in a multi-person, personal scene shows the character can carry a full scenario, not just solo shots.
The Takeaway
My repeatable method is three steps. First, build the character with a detailed style prompt that specifies rendering quality, lighting, and composition, not just the person. Second, lock that style so the model stops drifting. Third, change only the setting and action for each new scene. That is how you get one believable character across a whole course, which is what scenario-based learning actually requires. I built this entirely in Google Gemini, from generating the character to holding the style across every scene.