From manual to automated image editing: a paradigm shift.

Alex Ciorapciu
4 min readJun 3, 2020


There is no way of talking about image editing without mentioning Adobe Photoshop. It is the Alpha and the Omega. For 30 years it has been the omnipresent tool synonymous with image editing. And this is exactly the heart of the matter. Photoshop is a tool.

All the other apps out there for image editing, processing, retouching — they are all mere tools. Some are easy to understand, others come with documentation as extensive as a dissertation. It takes time to learn how to use them. Just as a sculptor needs to learn to use a chisel before he can create his first sculpture. To achieve a perfect cut out in Photoshop, you first have to learn how to use the magic wand tool, channel separation, anchor points for paths, etc.

There are many grandmasters of Photoshop who can create visual miracles. Their skill level took many years to build up. It’s the result of hard work, countless hours of practice and watching many, many online tutorials.

But the desire for good image editing does not only reside among these grandmasters. The owner of the bakery around the corner wants to advertise their birthday cakes without the oven in the image background. The marketeer wants the newsletter images in black and white with a large grain effect. The high-end fashion boutique wants the skin of their models airbrushed. Grandma wants a black and white image from her childhood recolored. I want my avatar picture to look like an 8-bit graphic from the ’90s.
We can all express what we want, we just can’t do it ourselves.

Example: 8-bit pixel art by

This is where the paradigm shifts. A new generation of services emerges, that are both a tool and a grandmaster in one.

In this new paradigm, you don’t need to operate the tool yourself anymore, the grandmaster does it for you. You only need to specify what you want. “I want the background removed.” It’s gone. “I want all skin moles to disappear.” They’re gone. “I want my image in color.” It is recolored.

Recoloring a photograph from black and white, depicting Thomas Edison and his second telegraph.
Example: Recoloring of a photograph by DeOldify

The magic that powers this revolution is called artificial intelligence. Or, to be precise, machine learning for computer vision. The new grandmaster of background removal is called semantic segmentation. The new grandmaster of recoloring is called a generative network. These “new grandmasters” are deep neural networks that learn from examples and are then able to replicate the learned techniques on images. They live in the cloud and feed of data and powerful GPUs that provide immense computational power.

Because you don’t even have to press that one infamous button, because new services replace manually operated tools, this new paradigm manages to achieve something unheard of: automated bulk-processing of images. While the old tools are all about single image editing, these new services scale. Once you tell them what to do, they can do that for a hundred, a thousand, or a hundred thousand images. Even better: it can be done as fast as you want or need it, as long as you can throw enough powerful machines at it. In fact, many powerful machines in the clouds are just eagerly waiting for their next calculation task.

Removing a skin mole in Photoshop is easy for a professional. Select the spot healing brush, tap the mole. Gone. That takes 1 second. With 20 moles? Well, 20 seconds. A hundred thousand images with 20 moles each could take around 23 days, granting you’re willing to work 24 hours per day, non-stop and don’t need to save your files in between. You get the point.

Bulk image processing of fashion images in
Example: Bulk image processing in autoretouch

But are the results as good as the ones from the old masters? Very close, but not yet. A machine makes mistakes, like humans do. The difference, however, is that once you improve the mistake of a machine, it won’t do it again. Moreover, the number of mistakes is, more or less, finite. The more images it sees, the less mistakes it will make. Therefore, it won’t be long until it will be better than the old grandmasters.

So is this the end of image editing by humans? No, it is not. Most of the works of the old masters contain an immense dose of creativity and skill. They’re not simple images, they’re art. It’s the tedious, repetitive work that is being automated.

A good analogy is photography. It can be an art and there are many highly skilled photographers that know how to do, for example, a white balance. Yet most of the images shot these days are created by you and me. On our phones. In fully automatic mode.

This democratizing effect that has forever changed photography is changing image editing, too. The baker, the marketeer, the boutique owner, grandma; they can all command this new generation of services, that are both tool and grandmaster in one. And they can do it now. All examples mentioned are already on the market and working on thousands of images as you read this. The paradigm shift has started.

Example: Chained up components in autoretouch make up a workflow that processes thousands of images.



Alex Ciorapciu