Multimedia learning is the process of acquiring and integrating knowledge from instructional representations that combine words and pictures. Words may be spoken or printed; pictures may be diagrams, photographs, graphs, animations, or demonstrations. The educational question is how these representations help learners explain a concept, retain information, and apply what they understand. A well-designed diagram can make a relationship visible while narration explains the sequence of events. Additional media can also introduce competing demands, obscure essential information, and leave learners with a convincing impression of understanding that later testing does not support. Research therefore evaluates what learners do with the representations and what they can do afterward, rather than treating the number of media formats as a measure of instructional quality (Mayer, 2024; Noetel et al., 2022).
The most useful general conclusion is conditional: coordinated verbal and visual explanations often improve learning, especially when material is conceptually complex and the presentation helps students identify relationships they would otherwise need to infer. Benefits depend on prior knowledge, pacing, task complexity, accessibility, the kind of graphic, and the outcome being assessed. Reducing irrelevant material, making correspondences apparent, and allowing learners to process essential steps are often productive; adding animation, narration, captions, or an instructor's face without a specific instructional function is less predictable (Cromley and Chen, 2025; Noetel et al., 2022).
What Multimedia Learning Means in Educational Psychology
Multimedia learning names a particular class of learning experience and the theories and experiments used to understand it. An illustrated explanation of the water cycle, a narrated mechanical animation, a graph with an accompanying interpretation, and a biology diagram embedded in a digital textbook can all be multimedia instruction. The definition refers to the informational combination of verbal and pictorial representations; it does not require a screen, internet access, simultaneous audiovisual playback, or advanced technology. A printed science textbook with explanatory images can be a multimedia learning environment (Mayer, 2024).
A productive representation contributes to the target understanding. A graph can reveal an interaction between variables, a labeled drawing can locate a mechanism, and a sequence of frames can depict changes over time. Decorative photographs, animation loops, and background music also occupy visual or auditory space, but their presence alone supplies no evidence that the intended concept has been learned. The distinction matters because learning materials should be judged against specific cognitive objectives, such as identifying parts, explaining a causal mechanism, applying a procedure, or reasoning with a new case (Mayer and Moreno, 2003; Mayer and Fiorella, 2014).
Multimedia presentation, learning, and performance
A learner can enjoy a clear animation, recognize familiar diagrams, and answer questions immediately after a lesson while still having difficulty recalling or applying its content several days later. Researchers distinguish factual retention, comprehension, inference, and transfer: retaining a term, explaining why a process works, interpreting an unfamiliar example, and applying the principle in a changed situation are related but different achievements. A visually polished lesson may improve attention or subjective satisfaction without improving each achievement to the same degree. Studies comparing instructional formats therefore need appropriate tests and, when durable learning is claimed, delayed outcomes (Cromley and Chen, 2025; Beege et al., 2023).
Historical Foundations and the Cognitive Theory of Multimedia Learning
The contemporary research program grew from cognitive accounts of verbal and visual processing, cognitive load theory, and experiments on how people learn from explanatory text and pictures. Richard E. Mayer developed the cognitive theory of multimedia learning through successive studies of how students select, organize, and integrate incoming information. The theory has evolved over several decades while retaining three central assumptions: partially differentiated verbal and pictorial processing channels, limited immediate processing capacity, and active construction of an integrated understanding (Mayer and Moreno, 2003; Mayer, 2024).
The theory proposes that learners attend to relevant words and images, organize selected material into coherent verbal and pictorial representations, and connect these structures to each other and to prior knowledge. This is an explanatory model of cognitive processing, not a claim that each sensory format automatically occupies an isolated compartment or that the brain has two limitless independent storage channels. The model predicts that design choices influence which mental operations students can carry out with the processing resources available at that moment (Mayer, 2024).
Dual channels and limited capacity
Visual material requires processing of shape, position, orientation, and spatial relations. Spoken explanation requires temporary handling of a changing linguistic signal. Written words enter visually and must also be interpreted linguistically, so a screen containing diagrams and dense captions can generate competition for visual attention. Presenting a relevant graphic with concise narration can sometimes distribute demands more effectively than presenting the same graphic alongside extensive printed sentences. This modality advantage is especially consequential when information moves at an externally controlled pace and the learner cannot conveniently look back (Ginns, 2005; Mayer and Moreno, 2003).
Working-memory limitations are central, but the amount of information that can be handled varies with how familiar the material is and how its elements interact. A beginner studying a complex electrical circuit may need to coordinate many unfamiliar symbols. A practiced engineer may recognize familiar configurations as meaningful units. Instruction that clarifies relationships for the beginner can become repetitive for the expert, a phenomenon studied as expertise reversal. The educational problem is to match the representation to the learner's current knowledge and the task's demands (Kalyuga et al., 2003).
Essential, extraneous, and generative processing
Essential processing concerns the cognitive work required by the subject itself: understanding how a neuron transmits a signal or how a mathematical proof progresses. Extraneous processing arises when the presentation imposes avoidable demands, such as searching for a legend far from the chart it explains, decoding unnecessary decoration, or reconciling narration with unrelated visuals. Generative processing describes the learner's effort to make sense of the essential ideas by organizing and integrating them. These categories guide design choices, although the processes are theoretically inferred and cannot simply be read from a single self-report score (Mayer and Moreno, 2003; Mayer, 2024).
Consider a short lesson on the movement of blood through the heart. Students need to understand chambers, valves, and the direction of flow. Their essential task is to coordinate those relationships. An unlabeled animation with rapidly moving colored arrows may leave students searching for which chamber is which. A clearly labeled diagram, a small sequence of annotated stages, and spoken explanation at an appropriate pace can free capacity for building the mechanism. Asking learners to predict where blood goes next can then turn available capacity toward active explanation. Such examples express a design hypothesis; the particular lesson still requires evaluation with its intended students and outcomes (Mayer and Fiorella, 2014; Wang et al., 2023).
What the Cumulative Research Actually Shows
Multimedia research is experimentally rich, but it has never established that every additional image or audio track improves learning. An overview of reviews by Noetel and colleagues identified 29 reviews encompassing 1,189 studies and 78,177 participants. It found meta-analytic support for several design principles, including contiguity, signaling, modality, and coherence, while showing that effects depend on instructional complexity and whether pacing is controlled by the system or learner. Because this is a synthesis of reviews, its study counts should not be interpreted as unique independent experiments across all pooled contrasts: reviews may overlap in their underlying studies (Noetel et al., 2022).
A 2025 meta-analysis by Cromley and Chen examined a more specific corpus: 92 peer-reviewed articles involving Mayer as an author, reporting 181 studies and 591 effect sizes from work published between 1990 and 2022. Its overall pooled effect was Hedges's g = 0.37. The mean differed substantially by instructional principle, media type, learning outcome, academic domain, age group, authorship, and other moderators. Some principles had clearly positive averages, whereas effects for several others were statistically inconclusive within that synthesis. Because the review samples one research lineage rather than every independent research team, its estimate should be read alongside broader reviews rather than treated as a universal magnitude for all multimedia instruction (Cromley and Chen, 2025).
These syntheses address partly different questions and sampling frames. Noetel and colleagues pooled prior reviews across broad multimedia-design literature, while Cromley and Chen investigated the boundary conditions of the Mayer research corpus. Different lists of statistically supported principles are therefore possible without invalidating the entire field. An average standardized difference also leaves major questions: Was the comparator an inferior slide deck or equally good text? Did the test ask for factual recall or novel transfer? Were the materials brief laboratory lessons or a sustained course? Was the effect replicated outside its originating laboratory? Such details determine what an effect estimate means for a teacher deciding how to teach a particular topic (Noetel et al., 2022; Cromley and Chen, 2025).
Evidence strength and interpretation
The broad proposition that relevant, coordinated words and pictures can support comprehension has substantial experimental and synthesis support. Evidence also favors careful placement of related material and attention-guiding signals in many contexts. The effectiveness of a specific technique, such as permanent subtitles on a narrated diagram or an instructor's talking head in a video, is more conditional. A useful evidence assessment separates a theory-consistent rationale from a demonstrated causal difference under a particular research design, and separates a statistically detectable difference from an educationally meaningful improvement that persists beyond the immediate test (Richter et al., 2016; Adesope and Nesbit, 2012; Polat, 2023).
The Multimedia Principle: Coordinate Words With Explanatory Pictures
The multimedia principle predicts that students often learn more deeply from words and relevant graphics together than from words alone. An illustration can externalize a spatial or causal relation that would otherwise require mental reconstruction from prose. A diagram of the phases of mitosis or a graph of interacting variables may make a relationship accessible that a verbally accurate paragraph leaves difficult to envision. This advantage depends on whether the graphic depicts the intended concept clearly and whether students can connect it to the explanation (Mayer, 2024; Noetel et al., 2022).
Pictures also introduce interpretive demands. Students must know what arrows, scales, legends, color coding, and schematic conventions signify. An illustration may simplify reality so aggressively that learners confuse the model with the phenomenon; an elaborate photograph may include detail irrelevant to the learning objective. A strong multimedia explanation tells the reader which relationships the image is meant to support, and helps distinguish observable evidence, model assumptions, and simplified teaching representations. The exact balance depends on the content and learner (Mayer and Fiorella, 2014).
Spatial and temporal contiguity
Spatial contiguity is the placement of mutually relevant verbal and pictorial information near each other. A short label next to the corresponding structure typically reduces visual search compared with a distant key requiring repeated back-and-forth inspection. Temporal contiguity concerns the alignment of an explanation with the visual event or state it describes. When learners must hold a sentence in mind while waiting for the related animation, they may have to perform additional coordination. The benefit is clearest when matching sources would otherwise be difficult to integrate (Noetel et al., 2022; Mayer and Fiorella, 2014).
In practice, place labels and annotations within the visual field they explain, synchronize narrated steps with the relevant image, and preserve the ability to review complex relations. A reader exploring a static diagram may benefit from persistent labels; a learner following a dynamic procedure may benefit from arrows or highlights that appear when needed. If the labels make the image crowded, progressive disclosure or a clearly linked legend may serve comprehension better than maximal proximity.
Coherence: Exclude Material That Competes With the Learning Goal
The coherence principle concerns avoidable content. Background music, decorative animations, tangential stories, and unrelated photographs may attract attention while failing to support the target explanation. Such additions can compete with relevant material for processing time or prompt mental connections that distract from the instructional objective. Research on coherence and seductive details has repeatedly shown that removing unnecessary information can improve performance in suitable tasks, although interest, prior knowledge, and the pedagogical purpose of an example all matter (Mayer and Fiorella, 2014; Cromley and Chen, 2025).
Coherence does not mean that every story, analogy, or emotionally engaging example should disappear. A concrete case may supply needed background knowledge or show why an abstract idea matters. A photograph can establish the context of an ecological problem. An analogy can connect an unfamiliar mechanism to a known system. The operative test is instructional: does the feature help the intended learner select, interpret, and connect essential concepts, or does it demand attention without adding explanatory value? Motivation and conceptual understanding should be evaluated rather than presumed from a designer's taste.
Signaling: Make the Structure and Relationships Visible
Signaling draws attention to relevant parts of an explanation through arrows, highlighting, headings, numbering, color correspondence, or verbal cues. It is especially useful where a learner must discover which object in a complex picture matches a sentence or where a process includes several interacting steps. A meta-analysis by Richter, Scheiter, and Eitel included 27 studies, 45 comparisons, and 2,464 participants. It found a positive small-to-medium association in the study's reported effect metric for comprehension and transfer, r = .17, with stronger benefits for learners with lower prior knowledge. The finding pertains to signals that highlight relationships between text and pictures and should not be generalized to every conceivable use of color or decoration (Richter et al., 2016).
Useful signals identify conceptual structure rather than merely increasing visual intensity. In a cell diagram, consistent color can connect a label, structure, and function. In a mathematical proof, numbering can reveal dependencies across steps. In a graph, a clear annotation can guide the reader to a change in slope. Too many highlighted elements abolish the hierarchy that highlighting was supposed to establish. Color must also be accompanied by labels, position, line styles, or other distinguishable cues so that the explanation remains interpretable to learners with color-vision differences or monochrome displays.
Modality: Decide What Should Be Spoken and What Should Be Read
The modality principle concerns the division of explanatory material between audible words and visible written words while learners inspect pictures. When a complex visualization requires substantial visual attention, concise narration may allow learners to follow the diagram without repeatedly shifting between a picture and a dense adjacent paragraph. Ginns's meta-analysis synthesized 43 independent effects and identified stronger modality advantages under system-paced conditions, with moderation by element interactivity and instructional domain. The practical advantage must be weighed against the costs and constraints of recording audio (Ginns, 2005).
The principle is conditional because spoken information is transient. Students who need to inspect unfamiliar symbols, replay an equation, translate terminology, review precise definitions, or work without sound can benefit from visible text. Printed labels may remain necessary even when the explanation is narrated. A self-paced reader can often coordinate a diagram and written explanation effectively by moving back and forth. Thus, choosing narration for an animated process should be a response to the task's processing demands, not a rule that all instructional text ought to be converted into speech (Mayer, 2014; Ginns, 2005).
Redundancy: When duplicating narration as text helps or hinders
Traditional formulations of the redundancy principle warn that simultaneously presenting verbatim spoken narration and large blocks of identical on-screen text can crowd an explanation and increase unnecessary processing. Yet research does not support an absolute prohibition on concurrent words. Adesope and Nesbit synthesized 57 independent studies and found that spoken-plus-written presentation outperformed spoken-only presentation under some conditions, while spoken-plus-written and written-only formats were not reliably different overall. Benefits depended on prior knowledge, pacing, and the presence of graphics; using concise key terms was often preferable to transcribing every sentence on the slide (Adesope and Nesbit, 2012).
A slide with a technical drawing, a short label for each component, and a separate spoken explanation may solve a different learning problem from a slide packed with text reproducing the narration word for word. Simultaneously spoken and printed language can help when comprehension of the language signal is itself difficult, as in second-language listening, though it may also overwhelm learners in a fast presentation. The critical variables include text density, speech rate, graphical complexity, learner language proficiency, and whether students can pause or reread (Montero Perez et al., 2013; Mayer, 2014).
Captions and accessibility require their own analysis
Captions serve access, comprehension, and linguistic learning objectives that are not exhausted by the conventional redundancy experiments. Students who are deaf or hard of hearing, those learning in a second language, and those who cannot use audio need usable alternatives to speech. In second-language education, a meta-analysis by Montero Perez and colleagues found benefits of captioned video for listening comprehension and vocabulary acquisition, with variation by outcome measurement. Such findings concern a particular language-learning context and cannot be assumed to apply with the same magnitude to every chemistry or geometry video (Montero Perez et al., 2013).
Educational design should make captions accurate, synchronized, readable, and controllable; provide transcripts when suitable; and preserve alternative access to information conveyed only by graphics. An automatic caption that misidentifies an equation or technical term can introduce errors while appearing to provide support. The learner's ability to turn captions on or off may help reconcile different needs. Accessibility is an obligation of inclusive educational provision, while particular learning-effect estimates belong to the populations, materials, and comparators actually studied. A universal claim that captions always improve learning, or always cause overload, misstates the evidence (Adesope and Nesbit, 2012; Mayer, 2014).
Segmenting and Pacing: Give Essential Steps Room to Be Understood
A continuous instructional animation can present a succession of dependent events before a student has organized the first. Segmenting divides the material into meaningful units that can be inspected before moving forward. In a worked demonstration, a segment might introduce the problem, another show the transformation, and another explain the resulting state. The goal is to support integration of a sequence, rather than break a lesson into arbitrary equal-duration clips. Allowing learners to pause, replay, and move backward can change the cognitive demands relative to an externally paced presentation (Mayer and Moreno, 2003; Noetel et al., 2022).
The evidence is more nuanced than the statement that shorter videos automatically cause more learning. In three experiments on segmented narrated lessons, Wang and colleagues found that segmenting alone improved retention but not transfer in an initial comparison; prompting learners to generate summaries during pauses improved retention and transfer in later comparisons. The study helps separate the effect of dividing content from the activity carried out during the pause. Segment boundaries provide an opportunity for processing; prompts can help learners use that opportunity (Wang et al., 2023).
Pre-training and managing unfamiliar vocabulary
Pre-training introduces the names and functions of essential components before asking students to understand a complex process involving those components. A novice watching a densely narrated animation of neurotransmission may struggle simultaneously to identify the synapse, vesicle, receptor, and transmitter and to trace the causal process. Introducing the parts first can reduce the number of unfamiliar relationships that must be handled at once. The strategy is most relevant when constituent elements are prerequisite to understanding the whole explanation. Its value should be assessed against the extra instructional time it requires (Mayer, 2024; Mayer and Moreno, 2003).
Pre-training is different from disclosing the entire answer in advance. Students can first acquire stable labels and simple functions, then use a richer multimedia example to explain relationships. For advanced students, extensive introductory definitions may be redundant. The key question is whether the prerequisite knowledge is genuinely absent for the intended audience.
Animation, Video, and Static Diagrams
Dynamic media are especially attractive when the target subject involves movement, transition, or a temporal process. An animation of how gears mesh or how a surgical movement is performed can convey a change that a single frame leaves implicit. A meta-analysis by Höffler and Leutner of 26 studies and 76 comparisons reported an average advantage for instructional animation over static pictures, d = 0.37, with stronger effects in some procedural and motor-learning contexts. The averages describe the included tasks and should not be treated as evidence that animation is superior for every concept (Höffler and Leutner, 2007).
For a stable structure, a static diagram may be easier to inspect and compare. For a temporal mechanism, an animation can reveal the sequence but may disappear before the learner identifies critical events. Key frames, pause controls, concise labels, synchronized explanation, and reviewable diagrams can combine the virtues of both formats. The relevant design choice concerns the information the learner needs to extract, the rate at which it appears, and whether the learner can recover it after it disappears (Höffler and Leutner, 2007; Cromley and Chen, 2025).
Does an on-screen instructor improve a video lesson?
An instructor's face can contribute social presence, perceived warmth, and attention cues. It can also compete with a diagram for visual space when the teacher's image lacks a specific function in the explanation. A systematic review by Polat examined 41 empirical studies and found mixed results on learning outcomes, with conditions under which social or attentional cues may be useful. Beege and colleagues' 2023 meta-analyses reported a positive average association with retention but not transfer, alongside changes in social and motivational perceptions. The evidence supports decisions based on the lesson's structure, not a universal requirement either to show or hide the instructor (Polat, 2023; Beege et al., 2023).
A teacher demonstrating a physical manipulation may need to appear on screen to make an action visible. In an abstract derivation, the equations and annotations may deserve most of the screen. Designers should compare learning under feasible layouts while measuring engagement separately from conceptual performance.
Generative Learning: Moving Beyond Passive Viewing
Multimedia materials organize the learning environment, but the student must still build knowledge. Asking learners to explain a step in their own words, predict a result, draw a causal diagram, compare cases, or answer a question can encourage deeper processing. Such activities should be calibrated: a novice asked to construct a full model without enough initial information may spend capacity guessing, while an experienced learner can profit from constructing connections independently. Multimedia design and active learning function together through the learner's interaction with the representation (Mayer, 2024; Cromley and Chen, 2025).
Pause-and-explain activities offer an example. After a short animation of heat transfer, a prompt might ask why the direction of energy flow changes in a new configuration. A strong question requests an inference that students can justify from the preceding representation. A weak question merely asks whether the video was interesting. In Wang and colleagues' experiments, generating summaries during the pauses was more effective for deeper learning than segmenting alone under their tested conditions. This supports making intervals cognitively purposeful rather than assuming pauses by themselves generate understanding (Wang et al., 2023).
Retrieval questions and attention during video lessons
Short retrieval questions inserted between lecture segments may help sustain task-related attention as well as encourage memory access. Szpunar, Khan, and Schacter conducted two experiments in which interpolated memory tests during online lectures reduced mind wandering and improved learning compared with the tested alternatives. These results support adding targeted questions when the goal is retention and focused attention; they do not establish a fixed optimal number of questions or replace the broader science of long-term retrieval practice (Szpunar et al., 2013).
A multimedia lesson can therefore present an explanation, ask students to retrieve or apply one essential relation, provide feedback, and proceed. The quality of the retrieval prompt matters: repeating a phrase verbatim is a different achievement from using a principle to solve a novel case. For the general mechanisms of memory testing and schedules of review, the encyclopedia's separate scientific treatment is Retrieval Practice: Why Testing Yourself Improves Memory.
Prior Knowledge, Expertise, and Learner Variation
Instructional effects differ across learners because their prior knowledge changes what must be processed. A beginner may need a labeled representation and spoken explanation to understand an unfamiliar system; an advanced learner may be able to inspect a compact diagram and generate the explanation independently. Kalyuga and colleagues reviewed evidence for expertise reversal: instructional supports that aid novices can become unnecessary or counterproductive as expertise increases. This argues for assessing prerequisite knowledge and adapting the amount of guidance as competence develops, rather than building separate immutable media formats for alleged learner types (Kalyuga et al., 2003).
Age, domain, and cultural or language familiarity also shape interpretation of pictures and words. A schematic familiar to a university engineering student may be opaque to a younger learner unfamiliar with its conventions. Second-language students can face extra demands in decoding narration and may benefit from subtitles or transcripts. Cromley and Chen's meta-analysis found moderation by age, academic domain, country or region, and other factors; such moderator patterns identify reasons to test generalizability, not deterministic rules for individual students (Cromley and Chen, 2025; Montero Perez et al., 2013).
Learner preferences are not verified learning-style matching rules
Some students prefer diagrams; others prefer listening or reading. Preferences are valuable information for comfort, access, and choice. They do not establish that matching instruction to a fixed visual, auditory, or kinesthetic category produces superior achievement. Multimedia learning theory investigates properties of tasks, representations, prior knowledge, and processing conditions. A student may need a spatial drawing to learn geometry, a sound sample to distinguish phonemes, and written symbols to examine an equation, regardless of a global declared preference. Offering accessible alternatives and allowing self-pacing is an educational-design decision that does not depend on diagnosing a learning style (Mayer, 2024; Noetel et al., 2022).
Inclusive and multilingual design
An inclusive multimedia lesson makes its informational relationships available through usable alternatives. Visual contrast, legible labeling, captions, transcripts, alternative descriptions, clear navigation, and controllable playback can help learners participate in the explanation. These features address access barriers. Their effects on measured comprehension depend on whether the alternatives preserve the relevant educational information, the learner's needs, and the cognitive demands of the task. A transcript that accurately conveys spoken explanation is useful; a transcript alone may not express spatial relationships that exist only in an image.
Designers should audit the semantic information conveyed by each medium. If a color change means “pressure has increased,” the same change should be named or represented by another cue. If a narrated demonstration references “this part,” its alternative description should identify the part precisely. For multilingual students, unfamiliar vocabulary should be made explicit and, when appropriate, connected to visual referents. Evidence on captioned second-language video illustrates how linguistic objectives can change the effects of combining text and sound (Montero Perez et al., 2013; Mayer, 2014).
How Researchers Measure Multimedia Learning and Cognitive Load
Randomized experiments commonly assign learners to alternative versions of a lesson: words alone versus words and graphics, continuous versus segmented animation, a signaled versus unmarked diagram, or narration with versus without captions. When assignment and implementation are sound, outcome differences provide evidence about the tested instructional contrast. Yet the comparator matters. A richly explained animation might outperform an impoverished text lesson because the animation conveys more essential information, while an equally informative static sequence could narrow the difference. The strongest interpretation isolates the property the experiment actually manipulated (Höffler and Leutner, 2007; Cromley and Chen, 2025).
Immediate retention tests ask for recently presented facts or steps. Transfer measures ask the learner to use the underlying principles in a new situation. Comprehension tests may require causal explanation, and process measures may include eye movements, viewing time, interaction logs, or ratings of perceived mental effort. Each instrument captures a different aspect of activity. Looking longer at a diagram is not automatically evidence of stronger comprehension; a lower mental-effort rating can mean either helpful reduction of extraneous demands or insufficient engagement with necessary material. Good evaluations combine outcomes and report exactly what each measure can establish (Mayer, 2024; Beege et al., 2023).
Why experimental effects may not generalize automatically
A short, controlled multimedia experiment can isolate a mechanism with precision, but daily classrooms involve interruptions, differing prior knowledge, accessibility needs, variable teacher explanations, unequal devices, and sustained curricula. Learning in a laboratory after one exposure may not predict independent performance weeks later. Many experiments use university samples and highly constrained topics. Broader syntheses reduce reliance on a single experiment, but variation among outcomes, participants, studies, and comparators remains. Reviews should be examined for sample coverage, publication bias, dependence among effects, and whether the analysis distinguishes learner age and domain (Noetel et al., 2022; Cromley and Chen, 2025).
A reported standardized effect is a comparison across study distributions, not a percentage increase in grades or a guaranteed benefit for each student. Positive pooled estimates can coexist with null or unfavorable results in particular contexts. Statistical uncertainty, heterogeneity, and the ecological validity of the treatment should guide decisions. An inexpensive improvement such as moving labels adjacent to a diagram may be reasonable to implement on strong theoretical and empirical grounds. An expensive immersive simulation should be subjected to a more demanding comparison against simpler instruction addressing the same learning goal (Cromley and Chen, 2025).
Practical Instructional Design: From Learning Goal to Tested Lesson
The first design decision is to specify what learners should be able to explain or do independently. If students must identify parts of a cell, a clean labeled image may suffice. If they must describe the sequence of molecular interaction, a short annotated animation may add useful temporal information. If they must compare competing explanations, placing graphs side by side with an explanatory prompt may better serve the objective. Media choices follow the conceptual and procedural structure of the material (Mayer and Moreno, 2003; Noetel et al., 2022).
Next, map prerequisite concepts. Identify unfamiliar terms, visual conventions, and intermediate steps students need in order to understand the main mechanism. Remove tangential details, place explanatory labels close to what they name, synchronize changing visuals and verbal descriptions, and use signals to make dependencies visible. Choose between text and speech after considering whether the visual field is already heavily loaded, whether information is transient, and what access alternatives are required. Preserve the ability to pause and review complex sequences where possible (Ginns, 2005; Richter et al., 2016).
Then add carefully positioned opportunities for learners to generate explanations or retrieve essential information. Feedback should clarify the relation between the learner's answer and the intended model. A well-timed prompt can expose misunderstanding before the next segment depends on that concept. It is useful to ask whether the lesson enables students to complete an immediate demonstration and whether they can later reconstruct the same principle without the demonstration. Durable learning may require practice outside the multimedia presentation itself (Wang et al., 2023; Szpunar et al., 2013).
Finally, evaluate the lesson against a defensible alternative. Compare equivalent content and time where possible, record who benefited, measure factual understanding and transfer separately, and collect accessibility feedback without confusing satisfaction with achievement. If a lesson is used in a course, include later assessments that require independent work. Iterative revision should respond to errors learners actually make: misreading a diagram, forgetting a crucial symbol, or failing to connect a moving image to its verbal account.
A worked example: teaching plate tectonics
Suppose the goal is to explain why earthquakes and volcanic activity concentrate around particular plate boundaries. First, a static world map identifies boundaries and major patterns, with readable labels and a legend whose symbols directly correspond to the map. Next, a simple sequence illustrates plates converging, separating, and sliding, with narration synchronized to the relevant arrows. A short pre-training segment introduces “lithosphere,” “subduction,” and “boundary” if these terms are unfamiliar. Decorative satellite imagery is included only where it explains a relevant landform or pattern. The design follows coordination, signaling, pacing, and prerequisite knowledge principles (Mayer, 2024; Richter et al., 2016).
After each mechanism, learners predict which geologic pattern would occur at a different boundary and justify their prediction. An immediate task checks whether they can identify the boundary types; a later unfamiliar-map task examines transfer. If students can name the three types but cannot interpret the new map, the next revision should strengthen causal integration rather than add spectacle. This is a hypothetical design example illustrating how research principles can guide a lesson; it is not a reported experimental result for plate tectonics.
A worked example: teaching mathematical procedures
A teacher introducing graph transformations can display a function and its graph, then change one parameter at a time while highlighting the corresponding graphical shift. The printed equation remains visible because students need to compare exact symbols. Speech explains the reason for each transformation without duplicating full paragraphs of on-screen text. Students pause between steps to predict the next curve and explain their reasoning. The lesson is evaluated against a comparably detailed static sequence, not a bare equation without instruction. Here the goal is to connect a symbolic representation to a spatial pattern, and the usefulness of animation depends on whether that moving representation clarifies the relationship for the intended learners (Mayer and Moreno, 2003; Höffler and Leutner, 2007).
Digital Platforms, Generated Media, and the Artificial Era
Digital systems make it easier to layer diagrams, speech, video, simulations, subtitles, and prompts. They also make it easy to overload a lesson with automatically generated animations, transcripts, and visual effects. The design principles remain questions about learning: which representation makes a relationship understandable, which competing demands should be reduced, and what evidence shows that students can use the resulting knowledge independently? Producing a more attractive or rapid explanation is not equivalent to demonstrating learning efficacy (Cromley and Chen, 2025; Noetel et al., 2022).
When generative AI creates a diagram, lesson narration, or quiz, educational review must also establish factual accuracy, correspondence between labels and images, correct sequencing, readable captions, and valid answer keys. A visually compelling but scientifically incorrect animation can support the learning of an incorrect model. These are editorial and instructional quality requirements derived from the goal of accurate knowledge acquisition; the multimedia evidence alone does not establish that generative AI tools improve grades, reasoning, or transfer. Evaluations of AI tutoring systems and automated homework support require their own controlled evidence and safeguards for student data.
Interactive systems can offer adjustable playback speed, alternative representations, and explanatory prompts, yet personalization adds real questions about accessibility, privacy, and whether learners retain independent control. Educational data should be collected only for defensible instructional purposes, with clear governance and attention to the consequences of erroneous profiles or automated recommendations. The relevant test remains how a tool changes meaningful, transferable learning under a credible comparator, including for students with fewer resources or less reliable access. The larger interdisciplinary educational context is covered in Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development.
Common Misinterpretations of Multimedia Learning
One mistake is equating multimedia with the simultaneous presentation of every available format. A screen carrying moving images, full narration, dense captions, decorative icons, and background music may increase processing demands without adding explanatory structure. More channels of presentation are useful when they clarify relationships that matter to the task; they are burdensome when they demand attention without aiding understanding (Mayer and Fiorella, 2014).
Another mistake is treating simplified design rules as laws without exceptions. The modality effect varies with pacing and content, redundancy effects vary with the presence of graphics and the learner's knowledge, and captions may serve linguistic and accessibility needs. Segmenting can help retention while failing to improve transfer unless pauses are used productively. The correct formulation of a principle specifies its mechanism, the contexts in which its effect has been observed, and plausible boundary conditions (Ginns, 2005; Adesope and Nesbit, 2012; Wang et al., 2023).
A third mistake is substituting attention, enthusiasm, or immediate fluency for durable understanding. Instructor presence and visual polish can improve some subjective experiences while measured learning effects remain mixed. A learner's confidence after watching a fluent explanation can exceed the ability to reconstruct it independently. Educational evaluation should include questions and applications that require the learner to produce knowledge, ideally after some delay (Polat, 2023; Cromley and Chen, 2025).
Multimedia Learning and Neighboring Scientific Concepts
Multimedia learning focuses on how coordinated verbal and pictorial representations can support educational understanding. Cognitive load theory analyzes how instructional demands interact with restricted working-memory resources, including element interactivity and the cost of presentation. The broader psychological study of mental workload also concerns demands outside education. The encyclopedia's distinct treatment of this general construct is Cognitive Load: How Mental Demands Affect Performance. The present article concerns the specific implications for words, pictures, audio, animation, segmentation, and instructional design (Mayer and Moreno, 2003; Mayer, 2024).
Metacognition refers to monitoring and regulating one's understanding; multimedia design can make conceptual structures clearer, while metacognitive activity concerns judging and adjusting what one knows. Retrieval practice strengthens access to learned information through attempts to remember it, and spaced repetition concerns the timing of review over longer intervals. These mechanisms can complement a multimedia lesson, but their independent scientific questions concern memory and self-regulation rather than the informational relationship between words and pictures. For longer-term study schedules, see Spaced Repetition: How the Spacing Effect Supports Long-Term Memory.
Frequently Asked Questions
Is multimedia learning better than reading a textbook?
A well-designed textbook containing explanatory figures is itself multimedia learning. Comparing a narrated animation with a printed textbook requires equivalent content, adequate time, and an outcome relevant to the task. Animation may clarify movement; a static diagram may allow more careful inspection; printed text may be easier to revisit. Broad reviews support several design principles, but do not establish an absolute rank ordering of learning media independent of content and context (Noetel et al., 2022; Höffler and Leutner, 2007).
Are pictures always helpful?
Pictures are most useful when they express information relevant to the educational goal and can be interpreted correctly. Decorative or confusing graphics can distract; specialized diagrams can require explanation and pre-training. A useful test is whether removing the picture would deprive the learner of an important relation or make it harder to reconstruct accurately (Mayer and Fiorella, 2014).
Should a video use narration and on-screen text together?
Use narration when it clarifies a visual process without overloading the visual field, and preserve essential labels, formulas, and accessibility alternatives as appropriate. Full verbatim text may be excessive in some formats, yet written-plus-spoken presentation benefits some learners and tasks. Empirical findings depend on pace, graphics, knowledge, and language demands (Ginns, 2005; Adesope and Nesbit, 2012).
What is the difference between cognitive load and cognitive effort?
Cognitive load concerns processing demands relative to available resources, while effort concerns how much of those resources the learner invests in the task. Removing an irrelevant distraction may reduce avoidable load; asking students to explain an essential principle may require productive effort. Lower self-reported effort is therefore not automatically a better educational outcome, particularly if learners have stopped engaging with difficult material (Mayer and Moreno, 2003).
Do short videos improve long-term learning?
Segmenting can help students process a sequence, but a short duration by itself does not guarantee understanding or transfer. Evidence from experiments on segmented lessons suggests that requiring students to generate summaries at pauses can add benefits beyond segmenting alone. Durability should be assessed with later testing and independent application (Wang et al., 2023).
Can AI-generated multimedia teach as effectively as human-designed lessons?
The answer requires controlled comparison of specific lessons, learners, tasks, and outcomes. The established research concerns features of instructional representations and cognitive processing; it does not establish a universal learning advantage for generated media. Any generated explanation must be checked for correctness, accessible presentation, and its effects on learners' independent reasoning. A fluent system response is evidence of content production, not proof of successful learning.
Scientific Conclusions
Multimedia learning research explains why coordinated words and explanatory graphics often support understanding, and why indiscriminate additions can undermine it. The most reliable design decisions follow the cognitive task: make relevant relations visible, align explanation with representation, guide attention to essential structure, manage transient information, and create opportunities for learners to explain and apply what they encounter. Adaptation to prior knowledge, linguistic needs, and accessibility is part of sound instructional design, not an optional embellishment (Mayer, 2024; Noetel et al., 2022).
Cumulative evidence supports a substantial field of experimental principles while revealing variation among methods, learners, media, and learning outcomes. The 2025 synthesis of Mayer's research corpus documents this variability explicitly. Educational practice advances when plausible cognitive mechanisms are tested against meaningful comparators and when learning is evaluated as retained, transferable knowledge rather than the immediate appearance of fluent performance. These requirements remain central as instruction expands into interactive and automatically generated media (Cromley and Chen, 2025).
Related Articles
Cognitive Load: How Mental Demands Affect Performance — General cognitive demands and mental workload across tasks, with links to educational cognitive load theory.
Retrieval Practice: Why Testing Yourself Improves Memory — The independent memory mechanism behind practice tests and active recall, complementing multimedia lesson design.
Spaced Repetition: How the Spacing Effect Supports Long-Term Memory — Scheduling review to support retention after multimedia instruction.
Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development — Wider educational questions raised by artificial intelligence, including assessment and independent learning.
Prior Knowledge and Misconceptions: How They Shape Learning — Examine how existing knowledge and misconceptions shape understanding, conceptual change, and instruction.
Metacognition in Education: Planning, Monitoring, and Learning — Explain metacognitive knowledge, monitoring accuracy, explicit instruction, and decisions about learning progress.
Transfer of Learning: Why Knowledge Does or Does Not Generalize — Investigate near and far transfer, abstraction, analogy, practice design, and limits of generalization.
Learning Styles: Evidence on Matching Teaching to Learner Preferences — Examine visual, auditory, and kinesthetic learning-style claims and experimental evidence for matching instruction.
References
Adesope, O. O., and Nesbit, J. C. (2012). Verbal redundancy in multimedia learning environments: A meta-analysis. Journal of Educational Psychology, 104(1), 250–263. https://doi.org/10.1037/a0026147.
Beege, M., Schroeder, N. L., Heidig, S., Rey, G. D., and Schneider, S. (2023). The instructor presence effect and its moderators in instructional video: A series of meta-analyses. Educational Research Review, 41, 100564. https://doi.org/10.1016/j.edurev.2023.100564.
Cromley, J. G., and Chen, R. (2025). A meta-analysis of Richard Mayer's multimedia learning research: Searching for boundary conditions of design principles across multiple media types. Educational Research Review, 49, 100730. https://doi.org/10.1016/j.edurev.2025.100730.
Ginns, P. (2005). Meta-analysis of the modality effect. Learning and Instruction, 15(4), 313–331. https://doi.org/10.1016/j.learninstruc.2005.07.001.
Höffler, T. N., and Leutner, D. (2007). Instructional animation versus static pictures: A meta-analysis. Learning and Instruction, 17(6), 722–738. https://doi.org/10.1016/j.learninstruc.2007.09.013.
Kalyuga, S., Ayres, P., Chandler, P., and Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. https://doi.org/10.1207/S15326985EP3801_4.
Mayer, R. E. (2014). Multimedia learning in a second language: A cognitive load perspective. Applied Cognitive Psychology. https://doi.org/10.1002/acp.3050.
Mayer, R. E. (2024). The past, present, and future of the cognitive theory of multimedia learning. Educational Psychology Review, 36, 8. https://doi.org/10.1007/s10648-023-09842-1.
Mayer, R. E., and Fiorella, L. (2014). Principles for reducing extraneous processing in multimedia learning: Coherence, signaling, redundancy, spatial contiguity, and temporal contiguity principles. In The Cambridge Handbook of Multimedia Learning (2nd ed.). https://doi.org/10.1017/CBO9781139547369.015.
Mayer, R. E., and Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38(1), 43–52. https://doi.org/10.1207/S15326985EP3801_6.
Montero Perez, M., Van Den Noortgate, W., and Desmet, P. (2013). Captioned video for L2 listening and vocabulary learning: A meta-analysis. System, 41(3), 720–739. https://doi.org/10.1016/j.system.2013.07.013.
Noetel, M., Griffith, S., Delaney, O., Harris, N. R., Sanders, T., Parker, P., del Pozo Cruz, B., and Lonsdale, C. (2022). Multimedia design for learning: An overview of reviews with meta-meta-analysis. Review of Educational Research, 92(3), 413–454. https://doi.org/10.3102/00346543211052329.
Polat, H. (2023). Instructors' presence in instructional videos: A systematic review. Education and Information Technologies, 28, 8537–8569. https://doi.org/10.1007/s10639-022-11532-4.
Richter, J., Scheiter, K., and Eitel, A. (2016). Signaling text-picture relations in multimedia learning: A comprehensive meta-analysis. Educational Research Review, 17, 19–36. https://doi.org/10.1016/j.edurev.2015.12.003.
Szpunar, K. K., Khan, N. Y., and Schacter, D. L. (2013). Interpolated memory tests reduce mind wandering and improve learning of online lectures. Proceedings of the National Academy of Sciences, 110(16), 6313–6317. https://doi.org/10.1073/pnas.1221764110.
Wang, F., Mayer, R. E., and Hu, X. (2023). Benefits of prompting students to generate summaries during pauses in segmented multimedia lessons. Journal of Computer Assisted Learning, 39(4), 1259–1273. https://doi.org/10.1111/jcal.12797.