Published October 29, 2026 12 min readBy the Character Animation AI team

Do animated characters help people learn? What the research says

A little, and only when the character does something useful. The best-known meta-analysis of on-screen “pedagogical agents” (Schroeder, Adesope and Gilbert, 2013, pooling 43 studies and 3,088 participants) found a small but statistically significant benefit for learning. Small is the important word: a character is one more thing on screen, and it either directs attention or competes for it.

Richard Mayer’s multimedia learning research is the best guide to which side of that line a character lands on. The principles are set out in his book Multimedia Learning (third edition, 2020) and, for characters specifically, in the social-cues chapter by Logan Fiorella and Mayer in The Cambridge Handbook of Multimedia Learning (2021). Five of them decide whether a character helps. In paraphrase:

Principle	What the research found	What it means for your character
Image	Putting the speaker’s or agent’s image on screen does not, by itself, reliably improve learning.	Presence is not the active ingredient. Never add a character just to fill a corner.
Embodiment	Learners do better when an on-screen instructor behaves in human-like ways: gesturing, moving, looking at the material, showing expression.	The character should move with purpose, not hold one pose or loop a dance.
Signalling	Learning improves when cues highlight how the essential material is organised.	Pointing, looking towards the content and reappearing at section breaks are cues.
Personalisation	Conversational wording (“you”, “let’s”) beats formal wording.	This is about the script, not the drawing. A character suits a conversational voice; it does not replace one.
Coherence	People learn better when extraneous material is left out.	A character bouncing beside dense text is extraneous. Take it off screen while the learner reads.

One study makes the embodiment point concrete. Li, Wang, Mayer and Liu (2019, Journal of Educational Psychology) compared an on-screen agent that pointed at the specific part of a diagram being narrated with agents that pointed vaguely, gestured without pointing, or did not gesture. Only specific pointing improved learning, and that group paid more attention to the relevant part of the screen. The gesture worked because it told people where to look at the moment it mattered.

Two caveats. Most of this research uses short lab lessons, often with university students, and the effects are modest, so treat the principles as strong defaults rather than laws. And the voice principle (a friendly human voice beats a machine voice) was established with the robotic synthetic voices of the 2000s; in 2017, Craig and Schroeder found that a modern text-to-speech engine and a recorded human voice both outperformed that older engine. Test narration on real learners rather than assuming.

The practical conclusion: a character earns its place by doing a job.

Five jobs an explainer or e-learning character should do

Before drawing anything, list the moments in your script where the character will act. If you cannot find at least three, you probably do not need one; a clear voice-over and good visuals may teach just as well.

Signal where to look. The character appears beside a diagram, chart or screenshot and points at the part the narration names, as it is named. It faces the content, never the edge of the frame, because learners follow gaze and gesture.
Set the tone. A compliance module and a product onboarding tour need different personalities. Choose one temperament (calm and precise, warm and encouraging, a little wry) and keep every clip in it.
Mark transitions. The character waves in at the start of a module, reappears briefly at each section break and leaves when the content takes over. That rhythm segments the lesson without extra words.
Give quiz feedback. A quick nod and clap for a correct answer, a sympathetic wince and shrug for an incorrect one, a thumbs-up before a retry. Feedback adds the most warmth for the least distraction, because the learner has just finished concentrating.
Model a behaviour. Raising a hand to ask a question, greeting a customer, pausing to think before answering. Short clips model a posture or a gesture well; a multi-step physical procedure needs real footage.

DecorationThe mascot bounces in a loop in the corner of every slide while the learner reads three paragraphs about data retention.

A jobAs the narration says “the retention period lives here”, the mascot appears, points at that field in the screenshot, holds for a beat, and steps out before the next paragraph.

One rule for personality: make the guide a peer, not an all-knowing expert. A character who reacts, encourages and occasionally hesitates is easier to like than a lecturer in cartoon form.

Build a character library by script beat

The method that makes this practical is to stop treating “the animated character” as one asset and treat it as a small library of clips, one per kind of moment in your script. Write the script first, mark each beat with a tag ([POINT-R], [CORRECT], [SECTION]), count the tags, and you have a shot list.

Every clip must show the same character, or the library falls apart the moment two clips share a lesson. In the studio that consistency is built in: you generate a design from a description or a reference image, then approve it, which locks the design so every later animation matches. Circulate the design to whoever owns the course’s look first; after approval you are directing motion, not redesigning. The how it works page shows both steps.

A starter library that covers most courses and explainers:

Beat	How to make it	Length
Hello / goodbye	Wave preset	4–6 s
Listening, waiting	Idle preset, set to back-and-forth in the editor so the loop never jumps	6 s
Point right	Custom: “Turns slightly and points to the right with an open hand, holds the point, then returns to a relaxed stance.”	4 s
Point left	Mirror the point-right clip in the editor (no credit)	4 s
Explain	Custom: “Gestures with both hands at chest height as if explaining a simple idea, calm and friendly, then returns to rest.”	6 s
Think, pose a question	Custom: “Taps chin with one finger, looks up thoughtfully, then raises eyebrows as if struck by an idea.”	4–6 s
Correct answer	Custom: “Nods quickly and claps twice, pleased, then settles back with a smile.”	4 s
Incorrect answer	Custom: “Winces gently, tilts head, then gives a sympathetic shrug and a small encouraging smile.”	4 s
Encourage	Custom: “Gives a warm thumbs-up and a small reassuring nod.”	4 s
Celebrate	Custom: “Pumps both fists in a short, happy celebration, then settles with a proud smile.”	4–6 s

Notes that save regenerations:

Keep the character in place. The camera does not move and the frame is square, so motion on the spot (turning, gesturing, nodding) works and “walks across the screen” does not. Do any travel in your authoring tool with a motion path.
End every clip at rest. “Then returns to a relaxed stance” lets clips cut into each other without a pose jump.
Give pointing a wind-up. A gesture with a clear start, hold and finish reads as deliberate; one that snaps into place reads as a twitch. The animation principles guide covers anticipation.
Check the mirror. Mirroring flips everything, including a logo on a shirt or a side parting. If the asymmetry will be noticed, generate a separate point-left clip.
Trim to the beat. The free editor trims (down to half a second), changes speed from 0.5× to 2× and reframes, so a 6-second take can become the 2.5 seconds the narration needs. Edited cuts are saved as versions.

The cost: one design plus nine animations is ten credits, with the mirrored clip and every edit free; allow a few more for retakes. The three free credits cover a design and two clips, enough to pilot a pointing clip and a feedback reaction in one real lesson.

Storyline, Rise, Captivate or Camtasia: which file to use

Every animation exports at once as transparent WebM, transparent MOV (ProRes 4444), transparent GIF (480 px wide, 15 fps, hard one-bit edges) and an opaque MP4 that the free editor can flatten onto any background colour you choose. Which to use depends on what your authoring tool reads with transparency. Here is what each vendor documents as of October 2026.

Tool	Best export	Transparency	Watch out for
Storyline 360	Transparent WebM (Insert › Video › Video from File)	Yes, since the December 2, 2025 update, in desktop Chrome, Edge and Firefox	Safari and most mobile browsers show a background colour. Publish with static video quality (adaptive publishes everything as MP4). MOV is converted to MP4 on import, losing the alpha.
Rise 360	MP4 with its background matched to the block, or the GIF in an image block	No alpha video documented	Rise re-encodes videos outside its recommended criteria to H.264 MP4, which has no alpha. For real transparency, embed a Storyline item with a Storyline block (same browser caveats).
Adobe Captivate 13	MP4 on a matched background, or the GIF as an image	GIF only, with hard edges	Captivate’s FAQ lists MP4 as its video format and GIF among its image formats; it documents no alpha video.
Camtasia	Transparent MOV (ProRes 4444)	Yes	TechSmith lists ProRes 4444 among Windows imports and MOV and GIF among its transparent media types. WebM import arrived in Camtasia 2025, but WebM transparency is not documented, so prefer the MOV.
Premiere Pro, Resolve, Final Cut, CapCut	Transparent MOV	Yes	Each editor has one alpha setting that matters; the video editor guide covers all four.
PowerPoint, Google Slides	Depends on the app	Varies	See the slides guide.

The Storyline row is the big change: Articulate’s WebM support article confirms alpha-channel transparency for WebM on many desktop browsers, so a soft-edged character can sit over any slide design. Two consequences: everyone who edits the project needs that update, because a project containing a WebM will not open in earlier versions; and learners on Safari, iPads and phones will see a background colour, so plan a fallback.

Matching the MP4 background is the most dependable trick here, and it is underused. If a Rise block or Captivate slide is a flat colour, set the same hex value as the MP4 background in the editor and the flattened clip looks identical to a transparent one on every device, soft edges included. It fails only over gradients, photos or anything that scrolls behind the character.

Test where your learners actually are. If much of the audience is on iPads or phones (field teams, retail, healthcare), use the matched MP4 for everyone, or the GIF for small placements. The transparent video formats guide explains why Apple browsers behave differently.

Placing and timing the character in a lesson

In Storyline 360

Confirm everyone on the project has the December 2, 2025 update (build 3.106.35838.0) or later.
Insert the WebM (Insert › Video › Video from File) and size it to a quarter to a third of the slide height, beside the content it points at.
On the timeline, start the pointing clip a few frames before the narration names the target. A late gesture reads as a reaction, not a pointer.
Put the correct and incorrect reactions on the quiz slide’s Correct and Incorrect feedback layers, so they play only when the learner answers.
Publish with static video quality, then test in Chrome and on an iPad before anyone else sees it.

In Camtasia and video editors

Drag the MOV onto a track above the screen recording, start each gesture slightly ahead of the line it supports, and fade it in and out over a few frames. Keep it clear of the cursor and any text the viewer needs to read. For a longer explainer, chain clips (Idle, a point, Idle, a celebrate) with short dissolves; because every clip comes from one approved design, the cuts read as one performance.

Rules that apply everywhere

Same side, same size, same height on every slide, so learners notice only what the character does. Adobe gives the same advice for Captivate’s avatars: place them consistently, never over important content.
Face the content. Use the mirrored clip rather than letting the character look away from the material.
Leave during dense reading. On a slide that is mostly text or a complex diagram, the character should step out.
One character per scene unless the script is a dialogue.

When you need a talking, lip-synced presenter instead

Be clear about what clip-based characters cannot do. The studio produces short, square, silent clips (4 to 12 seconds, up to 720×720) with no lip-sync or speech, no live puppeteering and no camera moves. They are gestures and reactions that sit beside your narration. If the design calls for the character to deliver the narration on screen, mouth in sync for minutes at a time, use a tool built for that:

Vyond is the common choice in corporate learning teams. Assign a mic recording, a text-to-speech clip or an imported audio file to a character and it lip-syncs, as Vyond’s help centre shows. It builds whole scenes too, and exports transparent backgrounds as WebM or GIF (MP4 turns them white).
Adobe Captivate now includes AI avatar narration: you write a transcript per slide and it generates a lip-synced presenter video, paid for with Adobe generative credits.
Adobe Character Animator computes lip sync from recorded audio for a puppet you build, if you want a hand-designed character and are willing to set up its mouth shapes.
Photoreal AI avatar platforms fit when the brief is a human presenter rather than a character.

Before switching, remember the image principle: a talking face does not automatically teach better than a voice-over with well-timed gestures. Adobe’s own best-practice notes for Captivate avatars suggest audio-only narration on slides with complex visuals. Many teams settle on a mix: a talking presenter for the welcome and summary, short gesture clips (or nothing) during the content. Gesture libraries are also cheaper to maintain, because rewriting a line of narration does not mean re-rendering a lip-synced take.

Accessibility: captions, distraction and motion

An accessible character is mostly a disciplined one. The checks that matter:

Caption the narration. WCAG success criterion 1.2.2 (Level A) requires captions for prerecorded audio in synchronised media, including meaningful non-speech sounds. The character clips are silent, but the lesson around them is not.
Never let the animation carry the only signal. A wince means “incorrect” only to someone who sees it and reads it that way. The feedback text must say so too.
Apply the five-second rule. Under WCAG 2.2.2 Pause, Stop, Hide (Level A), moving content that starts automatically, lasts more than five seconds and plays alongside other content needs a way to pause, stop or hide it. A four-second clip that plays once and stops stays clear; a looping idle character beside the lesson does not, unless the learner can pause or hide it.
Respect reduced motion. Where you control the web page, honour prefers-reduced-motion by showing a still frame instead. Authoring tools differ in whether they pass that preference through, so short, play-once clips are the dependable default.
Keep captions legible. Articulate’s guidance for transparent video recommends a background contrast of at least 4.5:1 for captions and controls.
No flashing in celebrations. Ask for “a short, happy celebration”, never strobing sparkles.
Hide decoration from screen readers. If a clip only reinforces the narration, mark it decorative so screen-reader users are not told about a waving owl on every slide.

Start with one lesson: a practical first week

Pick one existing lesson that has a diagram or screenshot and a short quiz. Mark the beats in its script.
Describe the character in words, or upload the mascot you already have, and choose a style. Flat sits comfortably beside most course templates; Soft 3D feels friendlier. The character design prompts guide covers how to describe one.
Get the design signed off by whoever owns the course’s look, then approve it in the studio.
Generate four clips: point right, explain, correct, incorrect. Mirror the point for the left side.
Trim each to its beat and export: WebM for Storyline on desktop, the matched MP4 for Rise and Captivate, the MOV for Camtasia.
Run the lesson with five learners. Ask what they remember, not whether they liked the owl.

Licensing. Internal staff training, like personal and educational projects, is covered on every plan, including the free credits. If you sell the course, build it for a client, or use the character in a company’s public marketing, you need the Professional plan, which is the only plan with a commercial licence. Details are on the pricing page.

A free account includes 3 credits once, with no card: enough for a design and the two clips that tell you whether a character belongs in your course at all. If it does, the library grows one beat at a time, always the same character.