Pixelmator Pro 1.3.4 adjusts the colors based on other photos

Pixelmator Pro with the new match color feature in the middle of the machine Learning is available on the Mac App Store. This is not the only novelty of this version 1.3.4 (44 €), but the most original.


The editor was given a taste of it a month ago showing that they were going to be able to extract the dominant colours of a photo and apply it to a second photo.


[embedded content]

Pixelmator Pro — ML Match Colors from Pixelmator Team on Vimeo



Practical example : you have a photo of a place that has been taken in the evening, and you want to get this vibe on the shot from another location in the middle of the day. All this without a complex manipulation, and in fact, it’s limited to a copy-and-paste or drag-and-drop. To do this, Pixelmator is based on the learning engine Core ML Apple with its own model.


In these first examples is simply to drag an image (the one taken at night) on the color settings of the target picture (in daylight). After a few seconds the software makes her work. The building to the left is not found in full darkness but one could think of a the end of the day.


Before

After

Before

The end result is not necessarily very realistic, but it was to regain a little of the atmosphere pop of the illustration to the right.


After

To make your tests, will load photos on Unsplash, there are a nice variety. Below is the starting image that we will change.


Original

Example 1 – source : Derick McKinney

Example 2 – source : Ash Edmonds

This also works with topics that we outline them in the starting image and that we are going to paste in the other image, hoping that the built-in element inherits the color tone. For that, we use the command “Paste and match colors ML” from the Edit menu.


The bug has been clipped (very quickly) and it does not jump, not necessarily the eyes, but it has also become automatically darker in the right image. The integration in terms of color, however, is not tip-top.

The tool has the merit of being very simple to use, with no settings. We find the principle of the filters while having a few hands on what you will be trying to get.


The developers of Pixelmator Pro, we had explained how they had trained the engine of this function, which is done by hand on 20 million of shots :



We have used images in the public domain. We first created an algorithm to select only those source images that are well exposed and well balanced, and exclude all those who did not respond to these qualities.


Then we created a model of machine learning as a separate “break” the images as naturally as possible (creating the classic problems of white balance and exposure). So we had $ 20 million of pairs of images. A bad photo that we had deteriorated automatically, and a photo original and correct.


Our model is trained to improve the images in the most natural manner possible, to obtain a white balance neutral, a balanced exposure; basically to give people the best starting point to get creative with their photos.



The publisher is the party of the framework Core ML Apple to design its own tool, that can be used on macOS as on iOS.


The other novelty concerns the functions of transformation of an image with the command “Transform” > “free Transform” from the “Organize” menu that allows you to change the perspective and angles in an image or change the dimensions of one of the layers.



Pixelmator is working now on version 1.4, which promised to be as important and “very oriented photos ” but in a way that will be original.

Comments