How can I detect AI-generated images and videos?
The AI-generated Image Detection model and the AI-generated Video Detection model determine whether an image or a video was entirely generated by an AI model, or if it is real.
How it works
The models only analyze the visual (pixel) content. No metadata is used, so stripping or altering EXIF, C2PA or other metadata has no effect on the result. Real photos and videos that have been edited, filtered or color-graded are not flagged as AI-generated.
The API returns an AI-generated score between 0 and 1, along with individual scores for each supported generator. The models cover the main image generators (Nano Banana, GPT-image, Stable Diffusion, MidJourney, Firefly, Flux, Ideogram...) and video generators (Sora, Veo, Runway, Pika, Kling...). New generators are added as they emerge. For videos, scores are returned for each analyzed frame, so you can locate AI-generated segments within a longer clip.
Scores above 0.5 generally indicate AI-generated content. For high-precision workflows such as identity verification or fraud detection, you can use a higher threshold such as 0.8.
AI-generated content vs Deepfakes
AI-generated content detection identifies whether the entire image or video was produced by a generative model. Deepfake Detection focuses on faces that have been swapped or manipulated inside an otherwise real image or video. Both models can be combined in a single API call.
Operations
Each image analyzed with AI-generated content detection counts as 5 operations, and each image analyzed with Deepfake Detection counts as 5 operations. For videos, the number of operations is multiplied by the number of frames analyzed.