Introduction
In light of the rapid advancements in AI-generative technology and its potential implications for the integrity of photographic competitions, the Canadian Association for Photographic Art (CAPA) has taken a proactive stance to address the use of AI-generated images and photos enhanced with AI-generative features.
This comprehensive document outlines CAPA’s position, guidelines, and the rationale behind the organization’s approach to safeguarding the authenticity and fairness of its photographic competitions.
Why AI-Generative Images Are Not Being Accepted In Competitions?
The decision to prohibit AI-generated images in photographic competitions stems from several critical concerns that challenge the core principles of photography.
- Unjustified Image Scraping and Corporate Profits – AI image generators often scrape internet images without proper compensation to the original photographer or artists. This practise raises concerns about the ethical implications of profiting from these unsuspecting image creators. By entering a prompt, the AI-generative created an image from elements contained in their AI dataset of scraped images.
- Copyright Challenges – The unresolved nature of copyright issues surrounding AI-generated images underscores the need for clarity in distinguishing between “intellectual property rights” and “fair usage” within the scope of AI-generative technology. This complexity adds legal challenges that photographic societies and associations must navigate before considering the acceptance of AI-generated content. In the United States, several AI firms are being sued for copyright infringement and the matters are before the courts.
- Copyright Authorship of AI Generated Images – Creators of AI-generative images are not receiving a copyright for their AI creations. United States Copyright Office’s “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” document outlines why countries are apprehensive in issuing copyrights to AI-generated images. Their guidance document states:
“Based on the Office’s understanding of the generative AI technologies currently available, users do not exercise ultimate creative control over how such systems interpret prompts and generate material. Instead, these prompts function more like instructions to a commissioned artist—they identify what the prompter wishes to have depicted, but the machine determines how those instructions are implemented in its output.“
“When an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.31 As a result, that material is not protected by copyright”
The majority of photographic societies and associations require that an entrant submitting an image into their competition must hold the copyright for their submitted image.
- Unfair Advantage – Accepting AI generated images in a photographic competition would create an unfair advantage, as AI algorithms are capable of producing images at a faster rate or with more precision than human photographers.
Due to the unresolved concerns highlighted above, the recognition of AI-generated images as legitimate photographs in our competitions is contingent on the resolution of these issues. Until then, participants submitting AI-generative images engage in fraudulent practices by presenting ineligible images for in photographic competitions, which is both deceptive and unethical.
The Majority of photographic societies and associations worldwide have adopted a firm stance, explicitly prohibiting the submission of AI-generative images and photographs enhanced with generative features. For instance, the Australian Photographic Society restricts such entries in their competitions:
Entrants must be the author of any image/s and all parts thereof submitted into an APS Competition.
All parts of the image or images must have been ‘photographed’ by the author.
‘Content-aware Fill’ or similar modification that is entirely based on pixels in the original image/s and which does not extend the image beyond its original boundaries is ‘acceptable’.
‘Generative Fill’ or other processes that use content generated by software from written prompts or developed from the work of others is ‘not acceptable’.
Question: “Can I use Generative Fill, or similar, to create or remove an object (person, flora or fauna, building, scenery effect etc) in my image?”
- This is not allowable if you use content generated by software from written prompts or developed from the work of others. For example:
- Removing an ‘unnecessary’ tree that is replaced with existing associated imagery is allowable.
- Removing a person and replacing them with a ‘dog on a lead’ is not allowable.
Removing the background of an existing bush scene and using text asking for a ‘beach background scene’ to be added is not allowable.
- Using generative fill to ‘add a waterfall’ to a scene where the waterfall is not part of an image taken by the photographer is not allowable.
- Turning a summer scene into a winter scene that introduces snow that was not part of an image taken by the photographer is not allowable.
- It is acknowledged that this is an area of rapid change and perhaps the most contentious”
Another example reflecting a comparable stance is The International of Federation of Photographic Art (FIAP), which serves as the umbrella photographic organization consisting of 94 photographic societies and associations worldwide:
“Info 260/2023 states:
Pictures created by artificial intelligence are not allowed in salons under FIAP patronage. Therefore, salon organizers are asked to include the following text into the regulations of their events:
Pictures created by artificial intelligence are not allowed in this salon! It is reminded that all parts of the image have to be photographed by the author who is holding the copyright of all works submitted. Offenders will be sanctioned for life!”
CAPA Proactive Approach To Address AI-Generative Technology
In the Fall of 2023, we updated our CAPA Eligibility to establish ‘guard-rails’ designed to address potential submission of AI-generated text-to-image and photos enhanced with AI-generative technology:
- For the purposes of our competitions, a photographic image is defined “as being a captured image on a light-sensitive device (e.g. film camera, digital camera, smartphone, tablet, etc…) and recorded on film or in a digital format.”
- An artificial intelligence (AI) generated image is not deemed to be a photographic image because it was created from scratch by the AI system and contains no image captured by the photographer. Therefore, this type of image will not be accepted into our competitions.
- Use of AI features contained within a post processing application (e.g. masking, sharpening, de-noise, enlarging, etc…) are permitted.
- Images that involve the use of AI generative techniques like in-painting (where AI generates pixels to fill in missing parts or removes and replaces selected element from the original photograph then using replace it with pixel element from the AI’s generative dataset) or out-painting (where the AI extends the image beyond its original boundaries by generating new elements) are not permitted for submission into CAPA competitions, regardless of whether text prompts were used or not.
- Potential winning images may be required to have supporting images (sky, texture, etc…which must have been captured by the submitting photographer) submitted on request by the Director of Competitions.
Furthermore, the Editing Criteria for all our competitions will be revised to include the following:
The Director of Competitions has the right to request and receive the original un-retouched JPEG or RAW file for a potential winning image in a competition for the purposes of verifying competition compliance.
Upon notification that their image is a potential winning entry, photographers may be required by the Director of Competitions to submit all original images, such as unretouched JPG or RAW files, along with other image files that were integrated into the submitted image.
Failure to comply with the Director’s request for image files will result in the potential winning image being withdrawn from the competition and the competition results will be re-sorted.
These requirements aim to ensure transparency and verify adherence to the competition’s specifications regarding image authenticity and compliance with the editing criteria.
Artificial Intelligence Definitions
In recent years, many photo post-processing applications have incorporated Artificial Intelligence features in their products. The emergence of AI-generative technology has created uncertainty among photographers regarding the acceptability of different AI features.
To bring clarity to the realm of artificial intelligence, the following definitions are provided:
- Artificial Intelligence (AI) – This a branch of computer science dedicated to simulating human-like intelligence.
- Machine Learning – A subset of Artificial Intelligence, machine learning employs algorithms and statistical models to enable computers to learn and perform specific task. Examples of AI machine learning features can be found in applications like Topaz Labs and On1 Photo Raw.
- Deep Learning – Another subset of Artificial Intelligence, machine learning involves the use of artificial neural networks to model and understand complex patterns and associated relationships. AI-generative technology is component of deep learning platforms. Examples of AI-generative apps include Midjourney, Open AI, Stability AI, Dream Studio, Night Cafe, Photoshop 2024’s Generative featured, Adobe’s Firefly text-to-image, Luminar AI’s generative features.
- Text-To-Image – This is a feature of AI-generative applications that and involves algorithmically generating images from a given text prompt. It eliminates the necessity for an original photograph and incorporates elements using a dataset sourced from images scraped from the internet, resulting in a realistic and seamless composition.
- Image In-Painting – This feature where AI generates pixels to fill in missing parts or removes and replaces selected element from the original photograph then using replace it with pixel element from the AI’s generative dataset.
- Image Out-Painting – This AI-generative technique extends the visual contents of an image beyond its original dimensions, offering an expansive view or a broader perspective.
- AI Rendering – The utilization of AI-generative algorithms and models to either generate new visual imagery or enhance an existing image.
Technology To Detect AI-Generated images
For all potential winning images, we utilize a dedicated AI classifier model that has undergone rigorous testing by myself, demonstrating a 99% reliability in detecting AI-generative text-to-image creations. This is AI machine learning model was trained on millions of examples, encompassing both AI-generated creations and human-captured images, and undergoes regular updates to ensure optimal performance with the latest generative engines. This AI classifier is not available as a free application and required a paid account.
Important to note, the AI classifier’s assessment is independent of the image’s metadata, providing a robust a comprehensive analysis. All potential winning images are subjected to this assessment.