People have a right to know when a realistic image, video or voice was made by AI. Labelling is also the law:
- India. The IT Rules, as amended in 2026, require services that let people create synthetically generated content to label it prominently, to embed permanent metadata or another technical marker including a unique identifier, and to stop users removing either.
- European Union. Article 50 of the AI Act requires AI-generated audio, images, video and text to be marked in a machine-readable way and to be detectable as AI-generated, and requires deepfakes to be disclosed.
- Platforms. Instagram, TikTok, YouTube and others ask creators to label realistic AI content, and many read content credentials automatically.
| Content | What you can see or hear | What’s inside the file |
|---|---|---|
| Realistic images: a real-looking person, or a scene that looks like a real place or event | A visible “AI-generated” label in the bottom-right corner | Embedded metadata, signed content credentials (with a trusted timestamp when available), usually an invisible watermark, and a unique generation ID |
| Realistic videos, including talking presenters, translated videos and film scenes, and videos with AI speech | The same label, on screen for the whole video | Embedded metadata and a unique generation ID; content credentials where the format allows |
| Audio in a cloned voice, and AI speech delivered as audio (for example from AI Voice Studio) | A short spoken disclosure at the start: “This audio was generated with AI.” | Embedded metadata and a unique generation ID |
| A Movie Director film with a line in a Voice Vault voice | A short spoken notice at the very start of the film, in the film’s language (or in English if it is in a language we have no recording for), with the same words on screen (in English: “This film uses AI-generated voices.”); each such line downloaded on its own keeps its own spoken disclosure | Embedded metadata and a unique generation ID; content credentials where the format allows |
| Videos from the AI Explainer Video Maker | The visible “AI-generated” label on every frame; when the video has a voice, a short spoken notice at the start that the audio was generated with AI (for a Voice Vault voice, see the next row) | Embedded metadata; content credentials where the file is small enough to be signed |
| An AI Explainer video narrated in a Voice Vault voice | A short spoken notice at the very start, in the video’s language (or in English if it is in a language we have no recording for; in English: “This film uses AI-generated voices.”), read by a stock voice and never by the cloned one, plus the visible “AI-generated” label on every frame; each narration recording also keeps its own spoken disclosure | Embedded metadata and a unique generation ID; content credentials where the file is small enough to be signed |
| Other AI images, video and audio: illustrations, logos, product-only shots, music and sound effects | No visible label (they don’t look like real people or events) | Embedded metadata, content credentials where the format allows, and a unique generation ID |
| Edits that don’t create new realistic content: background removal, upscaling, restoration | No visible label | Embedded metadata where the format allows |
| Text from our writing tools: posts, scripts, listings, summaries, song lyrics we write for you and documents | No visible label (text-only content is outside India’s synthetic-media rules) | A machine-readable AI mark in the formatted copy and in downloaded files, signed where we recognise the text (see AI-written text). Disclose it yourself where the law requires (see Your part) |
Some AI models add their own invisible watermark as well, for example Google’s SynthID in Veo and Nano Banana results. We never offer a way to remove any of these marks.
- The label reads “AI-generated”. It sits in the bottom-right corner, sized to the frame so it stays readable, and is part of the picture itself, not a layer that can be switched off.
- In a video it stays on screen from the first frame to the last.
- We decide automatically whether an output looks like a real person or a real event. When we’re not sure, we add the label.
The images, videos and audio files we generate on our servers carry information about how they were made, embedded in the file:
- the IPTC “digital source type” for AI-generated media (trainedAlgorithmicMedia), which many platforms and tools recognise;
- a unique generation ID, the date, “QikTools” and the name of the model used;
- where the format allows, a signed C2PA content credential, the open standard behind “Content Credentials”. Until our signing certificate is on the C2PA trust list, some checkers show our signature as valid but not trusted.
The metadata and credentials never include your name, email address or account details. They say how a file was made, not who made it.
PDF, Word, PowerPoint and Excel files and ZIP packs can’t carry Content Credentials the way pictures, video and audio can, so when one of our tools makes one from AI-generated material, we put a digitally signed AI mark in it instead.
- Where it lives. A PDF carries it in its metadata, a Word, PowerPoint or Excel file in its document properties (“AI-generated”, the generator, the date and a link to our verification page), and a ZIP pack as a file called QIKTOOLS-AI-PROVENANCE.json that lists every other file in the pack with its SHA-256 fingerprint and whether AI made it.
- What it says. That the file came out of one of our tools, on what date, how AI was involved (the IPTC digital source type trainedAlgorithmicMedia for AI-made material, or compositeWithTrainedAlgorithmicMedia where it is combined with other material) and a fingerprint of what we signed. It is a signed statement (a JSON Web Signature) made with the same certificate as our Content Credentials, whose certificate chain is public, so you can check it on our verification page, which also shows whether the file, or any file in a pack, has changed since.
- What it never contains. Your name, email address or account details, or the title or file name of a document. (A pack’s list does name the files inside the pack, because that is what it signs.)
- Your file stays with you. To sign it, your browser sends us only a fingerprint (for a pack: the names, sizes and fingerprints of its files). We never receive the file to sign it, and we keep nothing about the request.
- If signing isn’t available (you are offline or signed out, or our signing service is down), the download still goes ahead. A Word, PowerPoint or Excel file still carries its AI mark without our signature, a PDF keeps the AI note its tool wrote into it, and a pack’s list is marked “not signed”. The verification page then says “Marked as AI-generated by QikTools, not signed” (or that the list is not signed).
- What it doesn’t do. A signature says QikTools exported this file; it doesn’t say the content is true or checked, and a file that has been edited since shows as “changed”. Checkers that read only Content Credentials don’t see this mark; ours does.
- Files we don’t treat as AI-made aren’t marked as AI-generated: records, transcripts of speech, our own legal and compliance documents. Where a pack mixes AI-made and other files, its list says which is which (“generated”, “edited”, “mixed” or “none”).
Text from our writing tools — and lyrics we write for a song — is AI-generated text. A copy carries a machine-readable AI mark in its formatted (HTML) version, and a download carries it in the file itself: Markdown front matter, HTML meta tags and schema.org data, JSON entries, Word, PDF and PowerPoint properties, a note in WebVTT captions, or a labelled line at the end of a .txt file. Files that have no place for a mark, such as CSV and TSV spreadsheets and SubRip (.srt) subtitles, are not marked. We never put hidden characters or look-alike letters in your words. Where we recognise the words of a downloaded Markdown, text, HTML, JSON or WebVTT file as ours, we also sign its mark digitally. Word, PowerPoint and PDF files and ZIP packs are signed as described under Documents and ZIP packs.
Because plain text loses its details once it is copied, our servers keep a one-way fingerprint of each finished piece of about 30 words or more (never the text, and nothing about who made it) for 3 years, and “Check text” on the verification page compares a pasted text with those fingerprints. It finds exact copies, excerpts, lightly edited copies and generated passages inside longer texts; it can’t recognise text that has been heavily rewritten or translated, and “no match” never proves a person wrote something. The result says only whether, when and with which tool QikTools generated the text, never who made it.
We don’t mark small edits to your own writing (spelling, grammar, a word or two), transcripts of speech, code and formulas, alt text and other short labels (where a mark would change what a website or CMS imports), or, where a tool can tell that it’s yours, a translation of your own text. Lyrics you write yourself for a song are your own words and are not marked.
When you talk to Munns out loud you are talking to an AI: the panel says so before you start and during the conversation, and Munns introduces itself as QikTools’ AI assistant when it greets you. Its voice is AI-generated and is never presented as a real person’s. Nothing from the conversation is published or saved as content.
Anyone can check whether a file was made with QikTools on our verification page: upload the file or paste a link. If we find our credentials, it shows when the file was made and with which model, and that it is AI-generated. It never shows who made it, and the file you check isn’t stored.
Generated files are made by the models listed on the AI Models & Providers page, and the embedded credentials name the model used. Some of those models add their own marks as well. Some providers’ terms also let them use the content they receive to train their models; the Privacy Policy explains how each provider handles your content.
Never remove an AI label or content credential.
Don’t remove, crop out, blur, cover, alter or hide the visible label, the spoken disclosure, the metadata or the content credentials, and don’t re-encode, screenshot or edit a file to get rid of them, with our tools or anyone else’s. Removing them breaks our Terms, can break the law, and can lead to your account being closed.
- When you post realistic AI content, switch on the platform’s AI label where it has one.
- Say clearly that content is AI-generated where it could mislead people, and where the law requires. In the EU that includes deepfakes and AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and takes editorial responsibility for it.
- If you give QikTools results to your own customers or users, keep the labels and metadata intact and pass the disclosure on.
- Some websites and apps strip metadata when you upload a file. The visible label and our records stay.
- If a file can’t be marked in time, we still record its fingerprint, so it can be checked on the verification page.
- Files you export from our editors (Movie Director films, Video Studio clips, Design Studio and Carousel pictures and PDFs) are signed with content credentials when you export them, with our generation ID and, for most pictures, an invisible watermark. If signing can’t happen, you still get your file, with its visible label where it needs one. PDFs carry their credentials as a separate signed record linked from inside the file.
- Word, PowerPoint and PDF files from our document and writing tools (for example the Proposal & Slide Studio, the AI Resume Builder, the Ad Performance Lab and the Brand Kit) carry the signed AI mark described under Documents and ZIP packs instead of content credentials, and so do the ZIP packs our tools build. Design Studio, Carousel and storybook PDFs carry content credentials as a separate signed record linked from inside the file.
- A label says a file was made with AI. It doesn’t say whether its content is true, and no mark can prove who made or shared a file.
If you find QikTools content with its label removed, or content that misleads people, use the report form. For questions about labelling, write to hello@qiktools.com.