So you think you’ll get away with using Ai? Think again!

The most reliable way to detect if AI has been used in an image, for both total generation and generative removal (inpainting/object removal), is to use a multi-layered forensic approach rather than relying on a single tool.

Because AI models evolve rapidly, a combination of provenance standards, statistical pixel analysis, and metadata verification yields the highest accuracy.

1. Provenance and Cryptographic Tracing (Most Definitve)

The absolute most reliable "proof" comes from checking if the file contains hardcoded, tamper-evident cryptographic signals.

• C2PA / Content Credentials: Platforms like OpenAI, Adobe, and Microsoft automatically attach C2PA open-standard metadata to images.

  Even if an image undergoes "generative removal" (like Adobe Firefly's generative fill), a C2PA manifest will explicitly log that a generative AI tool was used on a specific region.

  You can verify this for free on the Content Credentials Verify Tool.

  
  

• Google SynthID: Developed by DeepMind, this embeds an invisible digital watermark directly into the pixel math. It is incredibly resilient and survives cropping, heavy compression, and screenshots. You can check for these signals using platforms integrated with Google's verification.

  
  

2. Statistical AI Classifiers (Best for Catching Un-marked AI)

If a bad actor has stripped the metadata or screenshotted the image to bypass tracking, you must rely on AI models trained to spot AI.

Generative removal leaves "boundary anomalies" and localized pixel changes that these platforms flag:

• Copyleaks AI Image Detector: Highly reliable for detecting "Mixed" content (images that are mostly real but used AI for object addition or removal) across models like Flux Fill and Midjourney inpainting.

  
  

• Hive Moderation: Widely used by trust and safety teams, scoring roughly 89–96% accuracy on raw generative files.

  
  

• TruthScan / DeepFakeDetector.ai: Top independent performers for catching deepfakes and localized Photoshop/AI tampering.

  
  

3. File and Error Level Analysis (ELA)

Generative removal creates a fundamental problem: the newly filled pixels have a different compression history than the rest of the original photo.

• Error Level Analysis (ELA): Tools like FotoForensics save the image at a specific failure rate and highlight the differences. If an object was generatively removed, that specific patch of the image will light up brightly or look completely different under ELA compared to the authentic background pixels.

  
  

• Frequency-Domain (FFT) Analysis: AI generators leave a "hidden order" or structural grid pattern in the spatial frequency of pixels. Advanced tools look at the Fast Fourier Transform (FFT) of the image to spot these mathematical signatures.

  
  

Summary Verification Workflow

To verify an image with near-100% confidence, process it through these four steps:

[Check Content Credentials] ➔ [Run through Hive/Copyleaks] ➔ [Run ELA for Localized Edits]

1. Provenance: Drop it into a C2PA viewer to see if AI usage is already legally baked into the file.

2. Triage: Upload it to Hive or Copyleaks to get a statistical percentage score.

3. Isolate: Use an ELA tool to see if a specific part of the photo has a mismatched compression layout (indicating generative removal).

At present, photographic competitions probably won't forensically check every single entry for Ai use, although scanning for illegal Adobe firefly use in the metadata of image files is currently trivial using batch C2PA searches and could process and flag 1000s of files in folder in seconds so this is something probably in the pipeline if not already being used.

Files wiped of their metadata will draw attention as this is a massive red flag as, why would you do this with nothing to hide? These images would probably be looked at on the pixel level by an Ai ELA tool to determine if areas have been removed or added using generative Ai.

So for anyone thinking “who's going to check thousands of entries for Ai” the answer is, Ai is going to check, and will be able to do it in a few seconds across competition folders where your images are stored.

“Ah! But it's a print competition” I hear you say!

Why do you think you are now often asked to submit PDIs of your printed image? It's not just for simultaneous projection and winners’ portfolios they can be used to check for illegal use of Ai too.

So, what to do?

Ask a search engine if the programme you edit with uses generative AI in its tools. Which tools, and can you disable the generative bit. Remember Ai IS allowed for editing if it's non-generative. This will be differentiated within a files' metadata so as not to be picked up in metadata searches for illegal generative Ai use.

“I don't want to have to start from scratch to re edit my submissions!”

In some programmes, with history stacks, you may be able to simply delete the offending edits, and fix using traditional editing methods like cloneing, without having to start from scratch.

If in doubt that you have used generative Ai tools, it's not worth submitting to major competitions like the NCPF, PSA etc as this will potentially get you personally banned or, worse, bring disrepute to your club.

The NCPF/PAGB have recently released warnings on the use of certain generative Ai tools in programmes like Photoshop, Topaz and Lightroom. I suspect this is so that clubs can not plead ignorance when illegal images are submitted and discovered going forward.

We have all been warned and as you can see from the tools they have for detection, there is nowhere to hide if you have used generative Ai.