AI tools

Learning goal

You will be able to sort the AI-based tools in editing programs according to what they do to a photo, and use each of them knowingly. You will explain the difference between tools that improve a photo's existing information, such as noise reduction and upscaling, selection tools that find the subject and the sky, tools that erase an object and fill its place, and generative fill, which produces the requested content from scratch from text. You will notice that these tools sometimes produce invented details, such as a hand, a piece of lettering or a facial line that doesn't really exist, and you will always check the result by enlarging it. You will grasp why a generated pixel doesn't carry a photo's value as evidence, which tools are restricted, and why, at the news desk, in documentary work and before a competition jury, and how to state honestly how an image was created with content credentials and an AI disclosure. Knowing that these tools may not work the same way on different skin tones, you will check the result on each face separately and complete the module's project.

The question

The band's social media manager suggests another way for the cover: instead of photographing the hall, typing a sentence into the program. When they type "a hall full of stage lamps behind four people," within a few seconds a hall that never existed, full of stage lights, appears behind the four members. The same program cleans up the noise in a photo shot at night with a single click, erases an unwanted stranger from inside the crowd, and enlarges a small photo to four times its size.

The drummer is impressed, but when they enlarge the generated hall they see strange things: a microphone stand on the stage melts halfway, the letters on a poster have turned into meaningless marks, and in the crowd, one person's hand has six fingers. In the night photo whose noise was cleaned up, the texture of the guitarist's dark skin has disappeared too. How is AI changing editing? Which tool improves a photo, and which turns it into something else?

Context

Throughout this module you have already met AI-based tools: automatic masks that find the sky and the person by themselves, content-aware fill that erases an object by filling from its surroundings. In the compositing lesson, too, we discussed what a combined image says in which context. AI tools sharpen these questions even further, because now it takes only a few seconds not just to combine existing photos, but also to generate images that were never shot.

These tools are changing fast; their names, menus and capabilities are renewed every year. That is why this lesson teaches not a particular tool, but sorting tools according to what they do to a photo. Once you can tell whether a tool improves existing information or produces new information, you can also decide yourself in which context it can be used.

Simple explanation

What do AI tools do to a photo?

An enhancement tool is an AI tool that makes a photo cleaner or larger while preserving its existing information. Noise reduction, blur removal and upscaling are in this group. These tools try to separate noise from detail, or add extra pixels to a small photo, using patterns they learned from a great many photos. Their results are often better than older methods, but they are still a guess: a noise reduction tool can mistake a fine texture for noise and erase it, and an upscaling tool can add a sharpness that doesn't really exist.

Selection and removal tools make up the second and third groups. Automatic masks don't change the photo's content; they only select an area. Removal and fill tools, on the other hand, remove an object and fill its place with content they guess from its surroundings; they are more powerful forms of the content-aware fill from the cleanup lesson.

Generative fill is a tool that fills an area not with the photo's own pixels, but with new content the AI produces from scratch, looking at a text request or at the surroundings. It can produce a sky, a hall, a person, an object. What it produces is an image no camera ever saw at any moment. That is why an area added with generative fill isn't a piece of a photo but is like a drawing or a painting; however realistic it looks, it isn't a record of that moment.

How do you notice an invented detail?

An AI hallucination is an AI tool producing a detail that doesn't really exist but tries to look believable. The places where it is seen most often are details that are familiar to the eye but hard to keep consistent: hands and fingers, letters and lettering, eyes and teeth, symmetrical objects, faces in the background. The lettering on a generated poster turning into meaningless marks, a hand carrying an extra finger, a microphone stand melting halfway are invented details of this kind.

Invented detail isn't seen only in generative fill. An upscaling tool can add eyelashes that don't really exist to a face, or brick lines that don't exist to a building; a removal tool can place a meaningless pattern where it removed a person. That is why every change made with AI needs to be checked not just by looking at the whole photo, but by enlarging the place where the change was made and going around its edges.

What is used in which context?

A photo's value often comes from its being a record of a moment. A generated pixel doesn't carry this value; it is evidence of nothing. That is why, in news and documentary photography, AI tools that add, remove or generate content in the scene are generally not accepted. Enhancement tools are often judged more flexibly, but they too carry the risk of inventing a detail. Competitions set different rules on this: some ban generative tools completely, some open a separate category, and some ask for every tool to be declared. Before entering a competition or giving a photo to a publication, you need to read the rules.

Content credentials are information carried inside the file as a record of which device an image was shot with and with which tools, and how, it was edited. There are standards developed for this purpose, and some cameras and editing programs can add this information to the file. An AI disclosure, on the other hand, is a note that clearly tells the viewer that AI tools were used in creating an image, especially that part of it was generated. It works on the same logic as the composite label: the viewer should know what they are looking at.

Depending on the diversity of the photos they learned from, AI tools may not work with the same success on different skin tones. A noise reduction tool can mistake the fine texture of dark skin for noise, erase it, and turn the face into a plastic surface; an automatic mask can make more errors at the edges of dark skin and curly hair. Generation and face enhancement tools, on the other hand, can lighten skin, change facial features or show the person differently from how they are while "correcting" a face. That is why you need to enlarge every face processed with AI and check whether it keeps the person's real skin, texture and facial features. Replacing a person's face with another face, or generating an image showing them doing something they didn't do, is not a matter of retouching but a much graver problem concerning the person's consent and dignity.

Visual evidence

The drummer puts the two versions side by side with the social media manager. When they enlarge the generated hall, they mark together the meaningless letters on the poster, the melting microphone stand and the six-fingered hand in the crowd. For the cover they settle on their own composite; with generative fill they fill only an empty corner of the hall's ceiling, and add the note "part of the background was generated with AI" to the cover's credits.

In the night photo they apply the noise reduction tool not at full strength but at fifty percent. The texture of the guitarist's dark skin and the roots of their stubble come back; the noise is still reduced, but the face isn't plastic. Then they rework from start to finish the close-up portrait they have been working on throughout the module: tone curve, local adjustments with masks, cleanup, measured dodge and burn, light tone evening with frequency separation, and a black-and-white version. They keep each step on a separate layer and name it, take a screenshot of each layer's before and after, and at no step generate a face with AI.

Common mistakes

Summary

AI tools are sorted by what they do to a photo: enhancement tools clean up or enlarge existing information, selection tools select an area, removal tools take an object away and fill its place with a guess, and generative fill produces new content no camera saw. These tools can produce invented details in places such as hands, lettering and faces, which is why every change is checked by enlarging it; a generated pixel isn't evidence of a moment, tools that change the scene aren't accepted in news and documentary work, and competitions set their own rules. Content credentials and an AI disclosure state honestly how an image was created; because the tools may not work the same way on different skin tones, each face is checked separately, and generations that change a person's face are a problem of consent and dignity.

Mini assignment

In the first three steps of this assignment you will test AI tools, and in the last step you will complete the module's project.

1. Apply an AI-based noise reduction tool to a noisy night photo at one hundred percent and at fifty percent strength; compare the two results by enlarging a face and a fine texture.
2. Remove a small object from a photo with a removal tool, and enlarge the gap to check whether there is an invented pattern or line.
3. If you can use it, add a small element to a photo with generative fill; enlarge the added element, mark the invented details, and write an AI disclosure when sharing.
4. Module project: retouch a portrait from start to finish. Work with the tone curve, masks, cleanup, dodge and burn and, if needed, frequency separation; keep each step on a separate, named layer, save each step's before and after state, and, with the person's consent, share the result together with its before and after states, explaining it step by step.

Check yourself

• Can you tell whether an AI tool improves existing information or produces new content?
• Were you able to enlarge a change made with AI and find the invented details?
• In your project portrait, did you verify at every step that you kept the person's skin, texture and facial features, and that you got the person's consent?

Glossary

Mini quiz

1. Which of the following adds new content to a photo that no camera saw?

• A) Noise reduction
• B) Generative fill
• C) An automatic sky mask
• D) The tone curve

2. Where are invented details in an AI-generated background seen most often?

• A) In a cloudless sky
• B) In the middle of the sky
• C) Always at the edges of the frame
• D) In details that are hard to keep consistent, such as hands, lettering, faces and symmetrical objects

3. What is the reason AI tools that change the scene are generally not accepted in news and documentary photography?

• A) A generated pixel not being a record of a moment, and the value of these photos coming from showing what actually happened
• B) These tools being expensive
• C) Their making the file larger
• D) Their changing the colors

4. When an AI-based noise reduction tool was applied at full strength, a dark-skinned person's face turned into a plastic surface. Which is the most fitting fix?

• A) Applying the tool once more
• B) Turning the face into black and white
• C) Applying the tool at a lower strength and checking, by enlarging the face, that the skin's texture is preserved
• D) Regenerating the face with generative fill

Answers

1. B — Generative fill produces content from scratch; the other tools work with existing information.
2. D — Details that are familiar to the eye but hard to generate consistently are the places most often invented.
3. A — A generated pixel is evidence of nothing; the reliability of these photos rests on their being records.
4. C — Moderate application reduces noise while preserving texture; each face must be checked separately.

Summary card

• Sort tools by their job: enhancing, selecting, removing, generating.
• A generated pixel isn't a record of a moment.
• Hands, lettering, faces: look for invented detail by enlarging.
• Don't change the scene in news and documentary work; elsewhere state it with an AI disclosure.
• Check each face separately; turning one face into someone else is a problem of consent and dignity.

Bridge to the next lesson

Having finished the module's project, the drummer wanted to get up from the screen and return to the moments when photos are shot. For the band's next music video they were going to shoot beside a stream; they wanted the flowing water in a photo to look like soft silk instead of frozen drops, and the crowd crossing the bridge to disappear from the frame without leaving any trace. This time it was a job not for the editing program, but for the camera, a tripod and a dark filter. How does water look like silk, how does a crowd disappear? The next lesson opens the advanced shooting techniques module with long exposure.