The State of Photo Editing: 2026 Research Report

Executive summary

Between January and April 2026, we analyzed anonymized aggregate data from 10.2 million photo edits across Magic Eraser's iOS, Android, and web platforms. The dataset spans users in 194 countries, though the majority of volume comes from the United States (34%), India (11%), the United Kingdom (8%), Germany (6%), and Brazil (5%). Edit types, session durations, device metadata, satisfaction ratings, and re-edit rates were captured. No individual images or personally identifiable information were included in the analysis.

Five findings stand out. First, object removal has solidified its position as the dominant edit type, accounting for over one in three edits. Second, mobile editing has overtaken web for the first time in our dataset, driven primarily by iOS growth. Third, AI-assisted editing is no longer a niche feature but the default workflow for most users, with three-quarters of all edits involving at least one AI-powered step. Fourth, e-commerce product photography is the largest single use case by industry, eclipsing social media and personal use. Fifth, speed improvements in AI processing have reduced average edit times by more than half compared to 18 months ago, at its core changing user expectations about turnaround.

Most common edit types

We categorized every edit into one of seven primary types based on the tool invoked and the scope of changes applied. Object removal leads at 34.1% of all edits. This includes removing people, signs, wires, trash, and other unwanted elements from photos. The typical object removal session involves selecting one to three objects for removal, with a median of 1.7 objects per session.

Background removal is the second most common edit at 27.8%. Users isolate a subject from its background for product listings, profile photos, design compositions, and social media content. Boost edits (brightness, contrast, sharpness, color correction) account for 17.6% of edits. While boost was historically the most common photo editing task globally, the rise of smartphone cameras with built-in computational photography has reduced the need for basic corrections.

The remaining categories are AI expansion or outpainting at 8.3%, creative fills and generative edits at 5.9%, text and watermark removal at 4.1%. Batch or multi-image operations at 2.2%. The long tail of edits that do not fit these categories accounts for the remaining fraction.

Platform usage trends

For the first time in our tracking history, mobile editing sessions have surpassed web sessions in total volume. Mobile now accounts for 63% of all editing sessions, up from 54% in mid-2025. iOS leads mobile usage at 38% of all sessions, with Android at 25%. Web-based editing has declined slightly from 41% to 37% but remains critical for expert and batch workflows where a larger screen and keyboard shortcuts provide meaningful efficiency gains.

The shift toward mobile is not uniform across edit types. Background removal and object removal skew heavily mobile (71% and 66% mobile, respectively), reflecting use cases like quick product photo cleanup and social media preparation. Boost edits are more balanced (55% mobile, 45% web), and batch operations remain overwhelmingly web-based (82% web). This suggests that users choose their platform based on the task complexity and the number of images involved, not simply based on device preference.

Geographically, mobile dominance is strongest in India (78% mobile), Brazil (74% mobile). Southeast Asia (76% mobile), while web usage remains comparatively strong in the United States (42% web), Germany (45% web), and Japan (43% web). These differences correlate with broader internet access patterns: mobile-first markets show correspondingly mobile-first editing behavior.

AI adoption in photo editing

The most striking trend in the dataset is the acceleration of AI-powered editing. In January 2025, 41% of edits on our platform used at least one AI-powered feature (object removal, background removal, AI expansion, generative fill, or AI-powered boost). By April 2026, that figure has reached 74%. The growth has been remarkably steady at roughly 2 percentage points per month.

AI-only workflows — sessions where every editing step is AI-powered with no manual adjustments — account for 39% of all sessions, up from 18% in early 2025. Hybrid workflows where users combine AI tools with manual adjustments (cropping, rotation, manual brush corrections) account for 35%. Fully manual editing has declined from 23% to 12% of sessions. The remaining 14% involves sessions where users open the editor but do not complete an edit.

Importantly, user satisfaction scores are highest for hybrid workflows (4.4 out of 5), slightly above AI-only workflows (4.3 out of 5). Notably above manual-only workflows (3.7 out of 5). This suggests that the combination of AI automation with human judgment currently delivers the best perceived results. Users who take an AI-generated edit and then fine-tune it manually report the highest confidence in their output.

Industry breakdown

Not all photo edits are personal. A major and growing share of editing activity on our platform is driven by expert and commercial use cases. We inferred industry categories from a combination of self-reported account types (for business accounts), edit patterns, and image content signals. The categorization is approximate and some overlap exists, mainly between social media and other categories.

E-commerce product photography is the largest single-industry segment at 31% of all edits. This includes sellers on Amazon, Shopify, Etsy, eBay. Other marketplaces who need clean product images with white or transparent backgrounds. The typical e-commerce user performs background removal and boost edits and processes an average of 4.7 images per session, greatly above the platform-wide average of 1.9.

Real estate photography accounts for 16% of edits. Agents and property managers use object removal to clean up interior and exterior shots, AI expansion to show wider room views. Boost to improve lighting in dark spaces. Social media content creation represents 14% of edits, with the fastest session times and the highest mobile usage share among expert categories. Personal use (non-commercial editing of personal photos) accounts for 22% of all edits. The remaining 17% is distributed across marketing agencies (7%), education (4%), journalism and media (3%), and other expert categories (3%).

Quality improvements and satisfaction metrics

Measuring the quality of photo edits at scale is inherently difficult because quality is subjective and context-dependent. We use three proxy metrics: user-reported satisfaction ratings (collected via optional post-edit surveys), re-edit rates (the percentage of edits where users undo or redo the AI result). Export rates (the percentage of sessions that end with the user saving or sharing the edited image).

Across all edit types, the average satisfaction rating is 4.2 out of 5, up from 3.8 in Q1 2025. The improvement is driven primarily by better AI model performance rather than changes in user expectations. Background removal shows the highest satisfaction at 4.5 out of 5, reflecting the maturity of segmentation models that now handle hair, fur, transparent objects, and complex edges with high reliability. Object removal satisfaction averages 4.3 out of 5, with scores varying by object complexity: simple objects (signs, wires, small debris) score 4.6 while complex objects (people in busy scenes, partially occluded items) score 3.9.

Re-edit rates have declined from 28% in early 2025 to 17% in Q1 2026. A lower re-edit rate indicates that users are satisfied with the first AI-generated result more often. Export rates have increased correspondingly from 71% to 83%, meaning more editing sessions now conclude with a saved output rather than abandonment. The combination of lower re-edits and higher exports suggests genuine improvement in output quality, not just user habituation.

Speed benchmarks

Processing speed directly affects user experience and workflow throughput. We measured median processing time (from the moment the user triggers an edit to the moment the result is displayed) across edit types and platforms. All times reflect server-side processing plus round-trip network latency for cloud-processed edits.

AI-powered object removal now completes in a median of 1.8 seconds, down from 4.7 seconds in Q1 2025 — a 62% improvement. Background removal is the fastest AI operation at 1.2 seconds median, benefiting from highly optimized segmentation models that have been refined over several generations. Boost edits complete in 0.8 seconds median, as many boost operations can be performed with lightweight models or traditional algorithms accelerated by GPU.

AI expansion and creative fill operations are the slowest, at 3.4 seconds and 4.1 seconds median respectively, reflecting the computational cost of generating new image content from scratch. These times are still well within the threshold that users perceive as responsive (research always shows that users perceive operations under 5 seconds as fast and operations over 10 seconds as slow for creative workflows).

Platform differences are meaningful. Web-based edits are about 15% faster on average than mobile edits for the same operation, primarily due to more consistent high-bandwidth connections. iOS edits are about 8% faster than Android edits on average, reflecting a combination of network infrastructure differences in the iOS user base and slightly more consistent device performance profiles among iOS devices.

Future outlook: trends for 2027

Based on the trajectory visible in our data and broader industry developments, we identify five trends likely to shape photo editing in 2027.

First, on-device AI processing will expand greatly. Apple, Google, and Qualcomm are all investing in neural processing units (NPUs) capable of running diffusion-based models locally. Our data shows that 6% of AI edits on supported devices already process on-device for simple operations. We project this will reach 15-20% by the end of 2027, driven by privacy preferences and the elimination of network latency for common edits.

Second, video editing will converge with photo editing. Our internal roadmap and public announcements from Adobe, Canva. Others point toward AI tools that handle video frames with the same ease as still images. Object removal from video, background replacement in video, and AI-powered video boost are all in active development across the industry. The user expectation established by instant photo editing will carry over to video.

Third, AI editing will become invisible. As AI processing becomes the default (already at 74% in our data), the distinction between AI editing and editing will fade. Users will expect intelligent behavior from every tool without thinking about whether AI is involved. This has implications for product design: the label AI will move from a feature badge to a background assumption.

Fourth, batch and workflow automation will grow. Our data shows batch operations at just 2.2% of sessions, but these sessions process disproportionately more images. As e-commerce and marketing use cases grow, we expect demand for automated pipelines that process hundreds or thousands of images against a consistent set of rules. API-based editing will grow alongside manual tool usage.

Fifth, quality expectations will continue rising. Each improvement in AI model quality resets user expectations upward. The 4.2 average satisfaction score in 2026 reflects higher absolute quality than the 3.8 score in 2025. Users are not simply harder to please, they are comparing against a better baseline. Maintaining satisfaction requires steady model improvement, not just maintaining current quality levels.

Methodology

This report is based on anonymized, aggregate telemetry data from Magic Eraser's production systems. The dataset covers 10,247,381 completed editing sessions between January 1, 2026, and April 30, 2026. A completed session is defined as one where the user invoked at least one editing tool, regardless of whether they exported the result.

All data was anonymized before analysis. No individual images were viewed or stored for this report. No personally identifiable information (names, email addresses, IP addresses) was included in the analysis dataset. Device and platform information was aggregated to the model family and operating system level. Geographic data was aggregated to the country level using GeoIP mapping applied during session logging, with the GeoIP data discarded before analysis.

Industry categorization was derived from three signals: self-reported account type for business accounts (available for about 23% of sessions), edit pattern clustering using k-means on feature vectors of tool usage, session duration, image count. Export format, and manual review of a stratified random sample of 5,000 anonymized session metadata records for validation. The industry labels should be treated as approximate estimates, not precise measurements.

Satisfaction ratings are collected via an optional one-tap post-edit survey presented to a random 15% sample of sessions. The response rate among those presented with the survey is 34%, yielding about 520,000 satisfaction data points. We applied inverse propensity weighting to adjust for response bias (users who complete high-quality edits are more likely to respond). Re-edit rates and export rates are measured from telemetry for all sessions and do not rely on surveys.

Speed benchmarks represent median and 95th percentile processing times measured server-side from request receipt to response dispatch, plus estimated client-side round-trip latency based on regional network performance data. Actual user-perceived times may vary based on device rendering speed and local network conditions.

This is the first edition of this report. We plan to publish updates semi-annually. Feedback on methodology and coverage can be directed to research@magiceraser.io.