Jump to content

Wikimedia Commons AI/Essay/January 2025

Add links
From Meta, a Wikimedia project coordination wiki
(English) This is an essay. It expresses the opinions and ideas of some Wikimedians but may not have wide support. This is not policy on Meta, but it may be a policy or guideline on other Wikimedia projects. Feel free to update this page as needed, or use the discussion page to propose major changes.
Translate
January 24, 2025 by S. Perquin

One year ago, I wrote an essay about Wikimedia Commons AI that I had envisioned (see here). Since it's becoming increasingly difficult to distinguish AI-generated media from human-created media (and I think that in the near future, it will also become impossible to recognize AI-generated media, see also simulacrumism), I believe it has now became even more important to establish Wikimedia Commons AI. Even though my proposal did not receive any support, I am still convinced of the importance and relevance my proposal.

On January 3, 2025, a user started a request for comment regarding AI-generated images and their place on Commons (see here). Wikimedia Commons AI can thus, in my opinion, provide a solution.

Dividing pictures into four categories

[edit]
An AI-generated conversation by NotebookLM about this idea of image classification into four categories.

After some thinking, on January 22, I came to the conclusion that it might be useful to create four categories for all images uploaded to Wikimedia Commons. (If Wikimedia Commons AI would be established, categories 1 and 2 would belong on Wikimedia Commons and categories 3 and 4 on Wikimedia Commons AI.) The exact definition and description of the categories has yet to be properly formulated. The categories I came up with are:

Category 1

[edit]
Works that are substantially unprocessed, unedited and untouched.

There has been no context-changing manipulation using photo editing software, artificial intelligence or other digital tools. So no sections in the image have been added or removed with, for example, a photo editing program. Examples are: unedited photographs, scanned photos and drawings made only by writing, drawing, painting or other physical materials.

  • This original painting has not been adjusted.
    This original painting has not been adjusted.
  • This is an original photograph in which no significant changes have been made.
    This is an original photograph in which no significant changes have been made.
  • No changes were made to this photograph either.
    No changes were made to this photograph either.
  • This is an original painting in which no edits have been made.
    This is an original painting in which no edits have been made.
  • This image was extracted from another file illustrated by humans. Because it has been cropped, it remains an original image because nothing has been altered.
    This image was extracted from another file illustrated by humans. Because it has been cropped, it remains an original image because nothing has been altered.

Category 2

[edit]
Works that have been manually edited and has or has not been processed by AI for only basic adjustments such as lighting or contrast.

This means there has only be made adjustments by physical or digital hand, such as adding, removing, editing, inserting or merging images. Excluded are technical aspects manipulated by AI, such as straightening, removing background, adjusting contrast, adjusting color levels, adding lightness/darkness, adding saturation, adding filters, perspective correction, etc. Examples are: digitally processed photos, a photoshopped image without the use of AI-generated content, human-illustrated images via a digital drawing or editing program, and physical collages made from cut-out non-AI-generated images.

Regarding scaled up or colored images by using or by not using AI, the issue is whether it is nicely done or not. If you see obvious flaws, such as eyes that are made too perfect compared to other parts of the image, they belong in category 3. If it is nicely done, it belongs in category 2. It is up to the users of the community to judge this.

  • A face mask was added to this painting by a human, without the use of AI. The image could be considered misleading, but since it does appear on Wikimedia Commons, it would belong under category 2.
    A face mask was added to this painting by a human, without the use of AI. The image could be considered misleading, but since it does appear on Wikimedia Commons, it would belong under category 2.
  • In this photo of the car, the license plate has been censored by digital software.
    In this photo of the car, the license plate has been censored by digital software.
  • This a photograph taken by a human and then also edited by a human, without the use of AI.
    This a photograph taken by a human and then also edited by a human, without the use of AI.
  • In this human made photograph, the background is blurred. It does not matter whether this was done by AI or not.
    In this human made photograph, the background is blurred. It does not matter whether this was done by AI or not.
  • This is a photographed portrait, but has been upscaled by AI. However, it does look good, so it can be seen as a basic adjustment by AI, which belongs in category 2. Opinions will be divided on this, which makes it difficult.
    This is a photographed portrait, but has been upscaled by AI. However, it does look good, so it can be seen as a basic adjustment by AI, which belongs in category 2. Opinions will be divided on this, which makes it difficult.

Category 3

[edit]
Works that have been relatively roughly manipulated or modified by an AI tool.

This includes everything in which AI-generated content can be seen, except what was originally generated by AI (they belong to category 4). This includes anything in which AI-generated content can be seen at all. Examples are: an AI-generated element added to a human-photographed work, an AI-expanded human-illustrated image, and a human-photographed work in which elements have been cut out by AI.

(Images under this category are hard to find on Wikimedia Commons. It could be that, for example, AI image extenders are not (yet) widely used, but it could also be that they are not yet correctly identified.)

  • An AI tool was used to artificially expand the background at the top of the image (to create more space above this person's head in this version). Since the top of the image was imagined by AI, which relied on other images it knows, this does not belong under category 2 but under category 3.
    An AI tool was used to artificially expand the background at the top of the image (to create more space above this person's head in this version). Since the top of the image was imagined by AI, which relied on other images it knows, this does not belong under category 2 but under category 3.
  • You can see from the eyes that this image has been upscaled by AI software. Something is not right, it looks off. However, opinions will be divided on this, which makes it difficult.
    You can see from the eyes that this image has been upscaled by AI software. Something is not right, it looks off. However, opinions will be divided on this, which makes it difficult.

Category 4

[edit]
Works that were initially generated by an AI tool and has either been modified or left unchanged.

This includes anything AI generated itself, regardless of whether and how it was modified by a human. Examples are: fully AI-generated images, AI-generated images photoshopped by a human using photo editing software, and an AI-generated image that is physically or digitally redrawn by a human.

  • This portrait was created entirely by AI.
    This portrait was created entirely by AI.
  • This image was also created entirely by AI.
    This image was also created entirely by AI.
  • Also this image was generated by AI.
    Also this image was generated by AI.
  • This image was generated by AI, and changes were also made to it, such as adding a second spray of water, by another AI tool.
    This image was generated by AI, and changes were also made to it, such as adding a second spray of water, by another AI tool.
  • This image was created by AI, but a human-made logo from Wikipedia was added to the image. That still makes it fundamentally AI-generated.
    This image was created by AI, but a human-made logo from Wikipedia was added to the image. That still makes it fundamentally AI-generated.

Responsibility

[edit]

You cannot control whether images are placed under the correct category by the uploader. People can consciously or unconsciously upload an image under the wrong category. My philosophy is that it is the responsibility of each individual to classify his uploaded picture in the category to which it belongs. For example, someone can redraw a unicorn generated by AI on paper or using a digital drawing program and then upload that work on Wikimedia Commons, but so can a copyrighted picture someone has redrawn. Furthermore, it is important that there are still people keeping an eye on whether images are under the right category.

Retrieved from "https://meta.wikimedia.org/w/index.php?title=Wikimedia_Commons_AI/Essay/January_2025&oldid=28180452"