Our mission is to automate professional level image editing. Over the past year, we've been working tirelessly to push the boundaries of what's possible in the world of real estate image enhancements. Today, we are announcing the beta launch of Autoenhance.ai 4.0 which will be available to test in our web app and via our API.
For 4.0 we went back to the basics, with a renewed focus on high quality image enhancements. As part of this process we expanded our development process with a new review step involving customers, photo editors and photographers to help score and provide feedback on our product. This provided the most comprehensive feedback process we've ever undertaken at Autoenhance, which helped us to identify the key areas for improvement. All coming together in Autoenhance.ai 4.0.
Improved Dynamic Range
We started with the core of our enhancement pipeline: the way HDR brackets are merged. Autoenhance.ai is able to handle all kinds of file uploads, from JPGs to RAW files and from single brackets to multiple brackets for HDR images. HDR images consist of multiple exposures of an image to capture as much exposure range as possible and has been the standard for professional real estate photography. These brackets then need to be blended to create an evenly exposed image with optimal dynamic range, preserving details in all areas of the image.
Our previous way of merging brackets would cause vignette effects in areas of intense shadows. We improved this in 4.0, which introduces a new technique for blending brackets that creates a more seamless transition between each bracket to get a more evenly balanced output. We also re-calibrated how our algorithm decides which areas are too dark and too bright to get an even better blend. And to top it all off we improved the decoding of RAW files during the HDR process to get much more detail and contrast. The end result is a merged image with high dynamic range, ready to be enhanced in the next part of our pipeline.
Global and Local Enhancements
In our quest to enhance an image just as a human editor does, we’ve been working closely with professional real estate image editors to learn from their workflow. Previous versions of Autoenhance.ai sometimes looked like a global filter had been applied to the image, where professional editors enhance local parts of an image differently. We’ve been able to recreate this workflow in Autoenhance.ai 4.0 by understanding how specific areas of an image need to be enhanced. It has learned to optimize saturation for one area, hues in other areas and so forth. So it’s not just making broad changes, but precisely calibrating the edits for each distinct area of the image. In addition to this, our new architecture allows us to more easily understand what edits the AI makes, which enabled us to create a completely new style preset. This new preset "authentic" preserves the look of your image as close as possible to the original, without making it look too warm or too cool.
Our new approach allows the AI to make both global and local adjustments as needed. For example, it might boost the overall contrast of the image, while also selectively improving shadows in one area and improving colors in other areas. The end result is a cohesive, professionally-edited look that's tailored to the unique characteristics of that particular image. And all of this has been trained on our extensive real estate datasets, so it has learned to optimize enhancements for real estate images specifically.
Window Pulls
High on our list of feature requests is the automation of window pulls. A common practice for real estate image editing is to mask out windows of interior images to improve exposure of that area. Human editors need to meticulously mask these windows and elements that are in front of the window, which is very time consuming. Editors then often blend in darker exposed brackets back into the window area to create a better overall exposure of the image. This has set apart human editors from AI solutions like Autoenhance.ai. Until today.
We built a custom neural network that has been trained on a vast dataset of real estate images including window pulls. By analyzing the visual patterns associated with window pulls across this dataset, the AI has developed a highly sophisticated understanding of where window pulls need to be applied. V4 is therefore able to scan new property images and accurately identifies the precise location of windows and masks them accordingly.
When a window pull is detected, the AI picks the best bracket which contains the most information for the window. It then intelligently blends this enhanced window pull back into the original image to create a professional-looking result. This level of automated image processing allows the window pulls to be elevated without looking artificially imposed, delivering a realistic representation of the room and its exterior. We believe this is a game-changer for automated real estate image editing.
Improved Auto Privacy
Previous versions of Autoenhance.ai already enabled customers to automatically blur license plates and faces, but there was room to improve accuracy and to add more elements. 4.0 is now not only more accurate but it can also blur for-sale signs, and picture frames that show humans.
When developing this feature we overhauled the way we trained our neural network. Expanding the dataset to thousands of examples specific to the real-estate market. It has learned to recognize the visual patterns that are indicative of humans in picture frames and for sale signs. Once the sensitive elements have been identified, the AI dynamically blurs these objects based on the size and location of the detected elements, ensuring a natural result without distracting from the rest of the image. On top of that, we have improved the accuracy of existing elements like license plates and human faces, helping our customers to ensure privacy even better.
Autoenhance.ai 4.0 builds on the already innovative features of our previous versions, like replacing grey skies and correcting perspectives. At the beginning of 2024, we welcomed two AI engineers to our team who, alongside the existing members, have diligently worked to elevate Autoenhance.ai to where it needs to be. 4.0 represents just a fraction of what Autoenhance.ai is capable of, and we have even more exciting developments on the horizon for this year.
4.0 Beta is now live in the Autoenhance.ai web app. Feel free to give it a try. We’re encouraging you to provide feedback if you encounter enhancements that need improvement. Your feedback is essential in our mission to enhance real estate images as good as a professional editor.