How to Distinguish AI-Generated Photos and Videos from Original Content
Jakarta (ANTARA) - The advancement of artificial intelligence (AI) is making photos and videos increasingly realistic, making them difficult to distinguish from original content. The public needs to be more diligent in checking sources, context, and digital footprints before trusting or sharing such media.
Generative AI technology is now capable of creating images and videos with extreme detail that closely resembles reality. Human faces, voices, locations, movements, and even specific environmental conditions can be created or modified using this technology.
This situation makes the ability to verify content authenticity increasingly important, especially when photos or videos are used to disseminate information regarding ongoing events.
Do not rely solely on visual appearance
One initial way to check a photo or video is to look for unnatural details. In several pieces of AI-generated content, anomalies can be seen in the shape of hands and fingers, text, reflections, shadows, facial expressions, or lip movements. However, these signs are becoming harder to detect as the quality of AI technology continues to evolve. Therefore, content that looks highly realistic does not automatically mean it is true.
Verify the content source
The next step is to find out where the photo or video originated. If a video claims to show a major event, check whether similar events are being reported by trusted media outlets or confirmed by official institutions. Pay attention to who first uploaded the content, when it was uploaded, and whether the accompanying caption matches the claimed event.
Photos or videos circulating only through anonymous accounts without clear sources should not be trusted immediately. For suspicious photos, reverse image searches can help determine if the image has appeared previously on the internet. This method is useful for finding old photos that are being reused with different captions or contexts. A photo claimed to be a recent event might actually originate from an incident several years ago. However, the failure to find a matching image does not automatically mean the photo is authentic; new or less widespread content may not yet appear in search results.
Observe metadata and digital footprints
Metadata can provide additional information about a file, such as the device used, the time of creation, or other technical details. This data can serve as a clue when examining a photo or video. However, metadata should not be used as the sole evidence, as information within it can be lost or altered when files are processed and shared through certain platforms. Verification can also be conducted by looking at the publication history and changes made to the content.
Recognise content credentials
An evolving technology designed to help verify the origin of content is ‘Content Credentials’. This technology allows information regarding the creation process and changes to a piece of content to be recorded digitally. Users can determine whether a photo was captured with a camera, edited with specific software, or involved AI. Content Credentials is one of the approaches being developed by the tech industry to provide more transparent information regarding content history.
However, not all photos and videos possess this feature, so the absence of Content Credentials does not necessarily mean the content is fake. Some technology companies are using digital watermarks to help identify AI-generated content. Google, for example, uses SynthID to embed a digital watermark that is invisible to the human eye into certain AI-generated content. This mark can be used to assist the verification process. By 2026, this technology has been expanded to assist in the examination of images and videos through specific services. Nevertheless, the absence of a watermark cannot be used as proof that content is an original camera recording.
Do not over-rely on AI detectors
Various tools are currently available that claim to determine whether a photo or video was created using AI. While these tools can be helpful, their results should not be considered absolute. Detection technologies have limitations and must constantly adapt to the developments in generative AI. A photo may even be a combination of original content and AI editing. For instance, an original photo of a person might be edited to change the background, remove objects, or add certain elements. Therefore, the question to be answered is not just whether a photo is ‘AI or not’, but also whether the photo has undergone alterations and to what extent those changes were made.
Be wary of videos featuring public figures
Videos featuring public figures should be examined more carefully, especially if they contain controversial statements, investment promotions, requests for money, or specific calls to action. Deepfake technology can make a person’s face and voice appear as though they are truly saying something they never actually said. If you encounter such a video, seek statements from the official accounts of the person involved or other trusted sources before believing it.
Another frequently overlooked factor is that the authenticity of a photo does not always mean the accompanying caption is true. A photo may indeed be taken with a camera, but used to depict a different event. For example, a photo of a disaster from several years ago may resurface and be claimed as a recent occurrence. In such cases, the issue is not a fake photo, but misleading context. Therefore, checking the date, location, and related events of a photo is just as important as verifying its visual authenticity. Do not immediately share suspicious content.