SINGAPORE, July 13, 2026 (GLOBE NEWSWIRE) — Airbrush today announced that its research paper, “Consistent Feature Transport for Image Relighting,” has been accepted to ECCV 2026, one of the world’s leading research conferences in computer vision. The paper has completed the camera-ready submission stage and is now moving toward official publication. The research is being applied to Airbrush Relight to help users create more natural-looking and visually consistent lighting effects in portraits and everyday photos.
Making AI Relighting More Natural
Poor lighting is one of the most common reasons otherwise strong photos look flat, dark, or unnatural. From dim indoor selfies and backlit portraits to social media posts and short-video covers, users often need more than a simple brightness adjustment. They need lighting edits that preserve facial details, skin tone, texture, and the overall balance between the subject and background.
AI relighting is challenging because changing the direction, warmth, intensity, or mood of light can easily affect the rest of the image. Some results may introduce muddy colors, unnatural shadows, uneven skin tones, blurred details, or a visible mismatch between the person and the surrounding scene. These issues become even more noticeable with complex lighting styles such as warm directional light, flash-inspired looks, rim light, or cinematic effects.
For creators, occasional success is not enough. A useful AI relighting tool needs to produce results that are not only visually appealing, but also consistent enough to use with fewer repeated generations and less manual correction.
A Research-Backed Approach to Relight
Airbrush’s CFT research approaches relighting as a problem of consistent lighting feature transfer. Instead of only learning what a well-relit image should look like, the method is designed to help AI models learn how lighting-related visual features can be transferred more consistently across different people, faces, and scenes.
This approach is designed to reduce common relighting issues such as skin-tone drift, subject-background mismatch, and loss of texture detail, while improving the overall visual coherence of the edited image.
The paper has been accepted to ECCV 2026 and has cleared the conference’s integrity and review-process checks. With the camera-ready version submitted, it is now moving toward official publication.
Bringing Advanced AI Relighting Into the Airbrush App
Airbrush is applying insights from the CFT research to improve its Relight feature, bringing research-backed relighting improvements into a simple, user-friendly editing experience.
For users in Airbrush App, the workflow remains easy: choose a photo, apply a lighting style, and adjust the look without needing professional lighting knowledge or complex manual retouching. Relight is designed for common creative use cases such as portrait enhancement, selfie editing, social media content creation, and short-video cover generation.
With these improvements, users can expect more natural-looking lighting effects, better preservation of facial details and texture, more consistent lighting between subject and background, reduced skin-tone drift, and less time spent regenerating or manually correcting inconsistent outputs.
In short, Airbrush’s CFT research helps move AI relighting from occasionally impressive results toward a more consistent and accessible creative experience. Instead of applying a simple filter effect, Relight is designed to help users reshape the lighting of a photo while keeping portraits natural and recognizable.
The full paper is expected to become publicly available following ECCV’s official publication process.
About Airbrush
Airbrush is an AI-powered photo and video editing platform that helps users enhance, retouch, and transform visual content with intuitive creative tools. Its features include portrait retouching, AI eraser, relight, background removal, video enhancement, and more.
Airbrush is available on iOS, Android, Windows, macOS, and web at airbrush.com.
A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/cc88bdb6-98b4-49a7-82fe-1fff243550bc
