FocusView: Custom Video Accessibility for ADHD

• FocusView is an accessibility-oriented system that customizes informational videos for ADHD viewers through multimodal channel adjustments.
• It enables viewers to personalize layout, background, captions, and audio to reduce distraction and enhance comprehension in educational and workplace settings.
• Empirical evaluation shows significant improvements in viewability and learning efficiency, underscoring the benefits of a customizable interface.

Searching arXiv for the cited FocusView paper and closely related context. FocusView is an accessibility-oriented video customization system for people with ADHD that treats informational video viewing as a configurable multimodal experience rather than a fixed stream. Its central premise is that videos are simultaneously appealing and distracting for ADHD viewers: multimodality, replayability, and stimulation can support attention, but the same properties can also amplify distraction. FocusView operationalizes this premise through a web-based interface that lets viewers customize layout, background, captions, and audio, with the goal of reducing distraction while preserving the informational value of the video (Zhu et al., 17 Jul 2025).

1. Scope, rationale, and target setting

FocusView is framed around informational videos rather than entertainment video in general. The motivating contexts are educational and professional settings, including educational videos, news, workplace training, and other material where attention and comprehension are task-critical. The system is presented as addressing a gap in both ADHD assistive technology and video accessibility research: prior work had examined distraction in reading, video conferencing, and data visualizations, but had not built tools for reducing distractions inside videos themselves for ADHD viewers (Zhu et al., 17 Jul 2025).

A defining design assumption is that distraction in informational video is multimodal and heterogeneous. The system therefore does not attempt a single universal correction. Instead, it decomposes a video into manipulable channels—speaker, content, auxiliary information, caption, background, and audio—and exposes a constrained set of customization options. This suggests a shift from “accessible video” as a fixed artifact toward “accessible video” as a viewer-specific configuration.

2. Empirical basis: distraction taxonomy and user needs

The design of FocusView is grounded in a formative study using in-the-wild social media data. The study collected the top 20 comments from more than 350 ADHD-relevant TikTok and YouTube videos, yielding over 7000 comments total. From comments where viewers indicated they had ADHD, the analysis identified four broad accessibility issues for ADHD viewers watching videos: prolonged video length, slow pace, missing captions, and distracting sounds and visuals. The first three were already partially addressed by existing platform features such as chapters, speed control, and auto-captions; the fourth motivated FocusView directly (Zhu et al., 17 Jul 2025).

The comment analysis also identified six potential distractors. These were the speaker’s appearance or behavior, content overlays, auxiliary information such as ads or unrelated inserts, background visuals, captions, and background audio. The important point was not merely that these elements could distract, but that they could do so independently and in combination. This led to a system architecture in which each category could be selectively modified rather than globally suppressed.

The subsequent user study reinforced a second, equally important finding: ADHD-relevant preferences were highly individualized. Background music, for example, could be experienced either as a distraction or as a stimulation boost. Captions could function either as an anchor for attention or as competing visual load. Background blur could reduce salience for some viewers while provoking curiosity in others. This heterogeneity is one of the paper’s central conclusions, and it underlies the system’s emphasis on customization rather than one-shot automation (Zhu et al., 17 Jul 2025).

3. Interface model and customization dimensions

FocusView is implemented as a web application built in React, with video processing on a remote server via FastAPI. The main interface presents a video player alongside four accordion sections labeled Layout, Background, Captions, and Audio. The accordion structure is deliberate: only one customization category is expanded at a time, reducing the number of simultaneously visible options and thereby reducing distraction caused by customization itself (Zhu et al., 17 Jul 2025).

The customization model is preset-based rather than unconstrained. The paper explicitly positions this as an ADHD-informed choice intended to reduce decision-making burden and sustained-task demands. The available controls can be summarized as follows.

Component	Options	Intended effect
Layout	Original, Auxiliary Removal, Speaker Focus, Content Focus	Rebalance speaker, main content, and overlays
Background	Blur or replace with a solid color	Dampen or remove distracting backgrounds
Captions	Color, font style, size, position, Dynamic Caption Tracking	Support reading and attention regulation
Audio	Denoise & Enhance	Remove background sounds and enhance speech

Layout customization is the most structurally consequential component. Speaker Focus enlarges and centers the speaker while removing content overlays. Content Focus enlarges and centers the main visual content, such as slides, charts, or code windows, while removing the speaker and auxiliary overlays. Auxiliary Removal preserves the speaker and main content but removes overlays deemed not directly content-relevant, such as watermarks, banners, or side graphics. Original leaves the composition unchanged. The paper treats these modes not as cosmetic presets but as attention-shaping transformations (Zhu et al., 17 Jul 2025).

Background customization targets busy or moving backgrounds. Viewers can blur the background or replace it with a solid color. The paper shows examples using white, dark gray, and peach replacements. Blur is intended to preserve some visual context while lowering salience; solid-color replacement removes the contextual background more aggressively. Caption customization uses VTT captions generated by Whisper and permits control over color, font style, size, and position. The offered color combinations are white on black, black on white, yellow on blue, and blue on yellow. The font options are Open Sans and a Bionic Reading-style font described as neurodiverse-friendly. Sizes are small (16px), medium (24px), and large (32px). Captions can be placed at the top or bottom and can also be dragged manually on the video screen. Audio customization bundles denoising and bandwidth enhancement using the resemble-enhance model (Zhu et al., 17 Jul 2025).

4. Computational pipeline and media decomposition

FocusView’s processing pipeline decomposes video into speaker, content, auxiliary information, background, captions, and audio. This decomposition is not purely conceptual; it is implemented through a sequence of CV and audio modules (Zhu et al., 17 Jul 2025).

Speaker segmentation starts with YOLO11 detecting all humans in the frame. The system then selects the most visually salient person as the speaker using TranSalNet. Once the speaker is identified, SAM2 performs finer-grained segmentation within the speaker bounding box. For presentation and overlay detection, televisions are detected with YOLO11. Other rectangular content, such as slides, code windows, or graphic panels, is detected with OpenCV using Canny edge detection and Probabilistic Hough Line Transform, followed by rule-based filtering to keep horizontal and vertical lines, form rectangles, remove tiny rectangles smaller than 5% of the frame, and merge nested ones. EasyOCR is used to find textual overlays such as watermarks. When removal is requested, holes are filled using the LaMa inpainting model. Background modification uses speaker and overlay masks to isolate the remaining frame and then applies Gaussian blur or solid-color replacement via OpenCV.

Layout classification into main content versus auxiliary information is rule-based. A detected element is treated as main content if it is a television; if it is a long-term overlay that appears consistently for more than 95% of the video duration and occupies more than 50% of the video frame in both height and width; or if it is a large, central overlay appearing in the middle third of the frame’s height and occupying at least 30% of the frame’s width or height. Anything else is treated as auxiliary information. This rule set matters because the semantic distinction between “main content” and “auxiliary information” is what makes the layout presets operational (Zhu et al., 17 Jul 2025).

Dynamic Caption Tracking is implemented by parsing VTT timing, estimating word timing by dividing segment duration by character count, and updating the highlight dynamically based on playback time. The highlighted current word is intended to support attention recovery and reading continuity. This suggests that FocusView’s caption subsystem is not merely decorative styling but a timing-sensitive attentional aid.

5. Evaluation design and empirical findings

FocusView was evaluated in a two-hour in-lab study with 12 adults with ADHD. Participants were 19 to 57 years old, with Mean = 29.7 and SD = 10.6. The sample included 7 females, 3 males, and 2 non-binary participants. Eleven had a clinical ADHD diagnosis; one was in the process of clinical diagnosis after being diagnosed by a psychotherapist. All participants watched videos daily. The study had three components after an initial interview: short video customization, long video segmentation/customization, and brainstorming discussion (Zhu et al., 17 Jul 2025).

In the short-video portion, each participant customized three short informational videos, one from each of three categories: educational, casual learning/talking-head, and news. The videos were all from YouTube, trimmed to under one minute, and chosen for diverse visual and auditory complexity. Participants first watched the first half in original form and rated its ADHD “viewability” on a 7-point Likert scale. They then customized the video in FocusView, watched the remainder in customized form, and rated the viewability again. The analysis used two within-subject factors, Condition and Video Type. Because the data failed normality according to the Shapiro-Wilk test ($p < 0.001$), the paper used an Aligned Rank Transform ANOVA, with post hoc ART contrast tests when needed (Zhu et al., 17 Jul 2025).

The main quantitative result was that FocusView significantly improved perceived video viewability, reported as $F = 165.4, p < 0.001, \eta^2_p = 0.75$. Participants also rated the system highly: effectiveness for reducing distraction and improving video watching was 6.17 (SD = 0.78), ease of use was 6.67 (SD = 0.44), and customization effort/workload was low at Mean = 1.54 (SD = 0.66). Qualitative responses tied these ratings to reduced rewatching and improved learning efficiency (Zhu et al., 17 Jul 2025).

The study also found that customization motivation depended on video type. Before customization, casual videos were more viewable than educational or news videos, with a one-way ART-ANOVA result of $F=9.20, p = 0.001, \eta^2_p = 0.46$. Post hoc ART contrasts showed casual videos were more viewable than educational ($t_{22} = 3.72, p = 0.003$) and more viewable than news ($t_{22} = 3.72, p = 0.003$), while educational and news did not differ ($t_{22} = 0.00, p = 1.00$). A plausible implication is that customization effort was perceived as more worthwhile when the video had academic or work-related stakes.

Preference data showed consistent diversity. Auxiliary Removal was the most frequent choice for educational videos (50%) and news videos (66.7%), and 11 of 12 participants used it at least once. Most participants who customized the background preferred blur rather than total removal, with 70% of background edits using blur. Nine participants turned captions on for all videos, one turned them off for all videos, and two toggled depending on context. Five participants denoised and enhanced all videos; three only did so for the casual video with music; two always kept original audio. These patterns reinforce the paper’s central claim that ADHD-relevant accessibility features are not uniformly beneficial or uniformly distracting (Zhu et al., 17 Jul 2025).

The long-video component, using two 3–4 minute videos each containing at least 10 scene switches, showed that viewers wanted both persistent defaults and scene-sensitive exceptions. All participants thought long videos should be segmented. Eight suggested having reusable presets, nine preferred ad hoc adjustments while watching, and three preferred making changes before watching. This suggests that a future version of FocusView would need not only per-video customization but also segmentation-aware control over when particular modifications apply.

6. Design tensions, limitations, and broader significance

FocusView’s most important design tension is the tradeoff between flexibility and cognitive load. Participants praised the fact that the system did not offer too many options, but they also requested more targeted controls, such as object-level background removal, adjustable blur strength, or more selective audio filtering. The paper treats this tension as ADHD-specific: too few controls leave distracting edge cases unresolved, while too many controls risk making the customization process itself into a distraction (Zhu et al., 17 Jul 2025).

A second major conclusion is that “less stimulation” is not a universal design rule. The study showed that some viewers wanted background music removed, while others described background audio as helpful stimulation. Captions could either support concentration or split attention. Background blur could either reduce salience or trigger curiosity about what was being obscured. This directly challenges any simplistic interpretation of accessibility as blanket sensory reduction.

The paper also identifies several practical limitations. The core hands-on evaluation focused on short videos with preprocessed customization variants, so it does not evaluate real-world waiting times for processing. The long-video part used a mock segmentation interface rather than full live customization. The study was conducted in a lab rather than in the field. The overlay detection approach focused on rectangular elements, leaving irregular or animated overlays as an open problem. Participants also raised concerns about processing delay, information loss, and the ethics of AI-generated replacement content. These caveats make clear that FocusView is best understood as a design probe and prototype rather than a finished deployment model (Zhu et al., 17 Jul 2025).

Within that scope, its significance is clear. FocusView demonstrates that informational video accessibility for ADHD viewers can be rethought as selective control over multimodal channels rather than passive consumption of a fixed audiovisual stream. Its strongest empirical claim is that this matters for perceived viewability; its strongest design claim is that ADHD-inclusive video accessibility must accommodate heterogeneous preferences rather than enforce a single notion of “reduced distraction.”