Synth ID Detector pricing: it is free

Yes. The bot has no paid plan and no credits: you send a file and get the answer.

Images as a file (JPG, PNG, WebP, HEIC), audio files, music, voice messages, video files and video notes. A photo sent with compression is not checked: the bot asks you to send it as a file.

When Telegram compresses a photo it strips all metadata, and the AI labels live in the metadata. As a file, the image arrives untouched.

Three ways. It reads the labels the generator writes into the file (C2PA, IPTC, XMP, PNG parameters, audio tags). For images it estimates the invisible SynthID watermark with a neural network. For audio and video sound it estimates an AI voice with another neural network.

Not necessarily. Labels disappear with screenshots, recompression and sending as a photo, and some generators never write them. The bot cannot see an unlabeled image from Midjourney, Flux, Qwen or Stable Diffusion, the picture of a video, or unlabeled music. Every answer without labels says so.

It is an estimate by an open approximation model, not Google's official detector. It finds about 9 in 10 Gemini images, and occasionally fires on other generators, so the bot calls the result likely. To check for sure, use synthid.com.

It finds about half of synthesized speech (ElevenLabs, TTS voice-overs, voice clones) and occasionally fires on a real voice. It does not check music or singing, and it does not judge recordings shorter than 4 seconds. For an ElevenLabs voice, use their AI Speech Classifier.

By the soundtrack and the labels of the file only. The bot does not analyse the picture and says so in the answer.

The bot downloads a working copy to run the check and deletes it afterwards. The check itself is stored as a short record of the file type, size and the result. See the privacy policy.

No. Synth ID Detector is a MadBots product and is not affiliated with Google. SynthID is the name of Google's watermark technology that the bot tries to find.