How does it work?

IsItAI does not run its own detection model. Instead, every image you upload is sent to Sightengine's "genai" API, a specialised AI-image-detection service, which returns a score estimating how likely the image is to have been generated or manipulated by artificial intelligence. This page explains, in plain language, what happens behind the scenes and what kind of AI content the model is designed to catch.

From upload to score, step by step

• 1. You upload a file. IsItAI accepts individual images (JPG, PNG, TIFF), PDFs and ZIP archives (which are unpacked so every image inside can be checked separately).
• 2. Each image is sent to Sightengine. The image is transmitted securely to Sightengine's servers, where its "genai" model analyses it.
• 3. A probability score is returned. Sightengine doesn't give a simple "yes/no" answer - it returns a probability (from 0 to 100%) that the image was produced or altered by an AI system. IsItAI turns that score into an easy-to-read verdict in your report.
• 4. Your report is generated. Once every file has been analysed, IsItAI compiles the results into a report you can view online, download (PDF/CSV) or receive by email.

What is actually being detected?

Under the hood, Sightengine's model is trained on very large sets of both genuine (camera-captured) and AI-generated images. Rather than recognising a specific software "signature", it learns the subtle statistical patterns, textures, lighting inconsistencies and compression artefacts that generative models tend to leave behind - patterns that are usually invisible to the naked eye but consistent enough for a classifier to pick up on.

This general approach means the model isn't limited to a single generator: it is built and regularly retrained to keep pace with the most widely used families of image-generation AI, including:

• Diffusion models, the technique behind tools such as Midjourney, DALL·E, Stable Diffusion and Adobe Firefly.
• GAN-based generators (Generative Adversarial Networks), such as StyleGAN, used for photorealistic faces and objects.
• Deepfakes and face-swap tools, which manipulate or replace faces in existing photos or videos.
• AI upscalers / inpainting tools, which partially edit or extend an otherwise real photo.

Because new generators appear constantly, no detector (including Sightengine's) can claim 100% coverage or 100% accuracy. Treat the result as a strong indicator to guide your judgment, not as absolute proof - and see our Sightengine Terms of Service summary for more on how these scores should (and shouldn't) be relied upon.

Why we use Sightengine

Building and maintaining a reliable AI-detection model in-house would require constant retraining against every new image generator released. By relying on Sightengine, a company dedicated full-time to this problem, IsItAI can offer up-to-date detection without you having to trust a black box built by us - Sightengine is a known, independent French provider with its own public terms and policies, linked from this site.

Curious about what data is shared with Sightengine to make this possible? See our cookie notice and the Sightengine Terms of Service summary.