Many online tools for detecting AI content are hosted on sites designed to assist people who want to create things using AI to make them appear more “human”.
Detecting Mixed Media is a problem: e.g. is a real photo modified with AI any different from a photoshopped one? Is a video manually edited with added emojis and cute animations any different from one tweaked with AI?
Music is a particularly complex subject.
How do you differentiate AI generated music from:
- Music generated by AI, but performed by a person e.g. this waltz we listened to in October: https://youtu.be/cLm9QUwGh4g?si=EauozPjVg9ISyT8z
- “Cover” songs – composed by humans but arranged and played using AI. e.g. This orchestrated cover of the hit song “Golden” https://youtu.be/YNyILzQ5_lQ?si=LRg7lUB8-i_dFMY1 Here’s the blurb from he creators : “We are a production team specializing in orchestral music, digital and real production, using the sound style of classical concerts performances, ai and hard work.”…
- Music composed and performed by a human but using a music synthesiser e.g. https://youtu.be/9uEp4GFDdII?si=tt_Y6CU3snl6PZtG
Watermarks in AI content
Google’s Gemini app now allows users to check if images were generated by Google AI, detecting Google’s invisible SynthID watermarks. While a step towards transparency and fostering content verification habits, this feature is limited to Google’s models and can be bypassed, offering only a partial solution for combating AI misinformation.
Even if no SynthID watermark is found, Gemini will attempt to analyze the image for tell-tale signs of AI. Gemini 3 Pro is particularly effective at detecting subtle AI artifacts, often spotting tell-tale signs the casual viewer might miss, such as remnants of common prompting terms appearing in text, objects with inconsistent physics, or an unnaturally smooth “AI skin” look. Beyond visual artifacts, it can also spot other clues, such as the file-naming conventions commonly used by generative AI apps and visible watermarks, like the ‘sparkle’ icon Nano Banana adds to its own AI-generated images. However, unlike SynthID, these techniques can’t confirm AI generation with absolute certainty. https://www.forbes.com/sites/paulmonckton/2025/11/23/can-you-tell-if-an-image-was-made-by-ai-google-just-gave-everyone-a-free-tool
It is worth checking the Google site: https://deepmind.google/models/synthid
SynthID is our new watermarking tool, designed specifically for AI-generated content. It empowers users to identify AI-generated (or altered) content, helping to foster transparency and trust in generative AI.
How SynthID works: SynthID embeds digital watermarks directly into AI-generated images, audio, text or video. The watermarks are embedded across Google’s generative AI consumer products, and are imperceptible to humans – but can be detected by SynthID’s technology.
In our tests, Gemini correctly identified a portrait taken with a TikTok filter, but shot itself in the foot when it went on to explain that it was an ageing filter – actually, the filter was supposed to be de-ageing !
Images and Video:
Here are some of the AI image and AI video detection sites we tested. Unfortunately, they tend to limit you to one use before you have to sign up. The results were patchy!
>>>Image detectors
https://sightengine.com/detect-ai-generated-images We used this a couple of years ago. It works well, although you have to sign up to do more than one test.
>>>Image and videos
https://matrix.tencent.com/ai-detect/ai_gen
Tencent’s Zhuque AI image and video detector can help identify whether an image or video is entirely generated by an AI model, or created by human. The detector is based on AI models trained on millions of AI-generated and human-created images and videos, covering a wide range of content such as photography, paintings, digital art, posters, movies, short videos and more. The tool currently supports detection of images and videos generated by mainstream models on the market, with more to be supported over time. This may work with Chinese AI content creators software, but didn’t spot the TikTok filtered image. It’s free to use up to a daily quota of 20 free text detections and 30 free image detections.
>>>Video detectors
Open source on Github – DeepWare:
https://scanner.deepware.ai
This failed to detect this video as being Deep Fake, although it is 100% AI generated! : https://youtu.be/KHSr80Rnm9k?si=YlEc1CppZRCNO27x
Music:
The BBC published an article on how we human beings might be able to spot a song generated by AI: How can you tell if your new favourite artist is a real person? https://www.bbc.com/news/articles/c5ylzjj5wzwo Chris remarked that many pop songs are formulaic, because that’s what people like.
>>> AI Music detectors.
Submit Hub:
https://www.submithub.com/ai-song-checker Upload an MP3 or Soundcloud Youtube Disco or Spotify links “I trained it on two types of audio files: AI-generated samples,Human-made samples To keep things balanced, I pulled an equal number of samples from each set (around 2,000 for each). That way, the model doesn’t lean too much one way or the other. Each audio file goes through a bunch of feature extraction steps, and that data trains the Random Forest Classifier. My detector checks 21 features across three main areas.
Note: You must sign up or log in before using this tool.
Let’s Submit:
https://letssubmit.com/ai-music-checker
Upload an MP3 or spotify link. Check if a song is fully or in parts AI-generated. This is an early beta version which we are training and improving regularly. We cannot guarantee 100% accuracy just yet… But we are getting there. Enjoy
This failed with a piece created by AI, but played by a human, but was successful with this song,which is entirely AI-generated : https://youtu.be/KHSr80Rnm9k?si=YlEc1CppZRCNO27x
Chris remarked that people say they are generating AI music so that they can have a personal playlist of stuff they like, rather than being bombarded with songs the music companies want you to listen to. Also, the songs we love are often associated with some memorable event in our lives. If a song raises the hairs on your neck – does it matter if it was created by AI or not?
Text:
Tell-tale signs can be spotted by humans, but the large language models are getting better. Wikipedia has a comprehensive section on how to spot AI generated text. https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
AI Checker : This is an app for detecting AI text which works with Google products:
https://workspace.google.com/marketplace/app/ai_detector_ai_checker/273593096182 . It needs to be installed first. Although it says “it has been added to Docs, Sheets, Slides, Drive and Gmail sidebar. Chris only saw it in Gmail – perhaps because she’s not paying for extra Google services.
Grammarly
https://www.grammarly.com/ai-detector
Navigate responsible AI use with our AI checker, trained to identify AI-generated text. A clear score shows how much of your work appears to be written with AI so you can submit it with peace of mind.
This works, but nags you to sign up with Grammarly
Quillbot
https://quillbot.com/ai-content-detector The free version is very limited
Textguard
https://textguard.ai This works, but you have to sign up to do more than one test. This site also has tools to create good quality text with AI, including a module called “Humaniser”,
NoteGPT AI detector
https://notegpt.io/ai-detector
Our AI Detector offers near 100% accuracy in detecting content generated by AI models like ChatGPT, Gemini, Claude, and more. Ensure authenticity with confidence. This correctly differentiated between text written by a human and that generated by AI. The site offers tools such as an AI paper writer and AI essay writer (at a price, of course)
AI Generated content is here to stay. So we have to get used to seeing it and ensuring that it is used transparently. The problem of copyright infringement is a very thorny one.
Christine Betterton-Jones – Knowledge Junkie.