AI labelling at Ikrames

Since August 2026, Article 50 of the EU AI Act applies. In short: AI-generated content must be recognisable as such. Part of that runs automatically at Ikrames, part is a setting, and part stays with you. This page separates the three.

What Ikrames does automatically

Every file created at Ikrames carries machine-readable provenance data — Content Credentials, following the C2PA standard. That covers generated images, carousel slides, the finished rendered reel, and the labelled version created at publishing time: it is re-marked after the icon is burned in, so the data matches the file.

This happens invisibly: you see nothing of it, and it cannot be switched off. The provenance data sits in the file container, alongside the pixels.

For images there is a second, entirely different marking: an invisible watermark computed directly into the pixels. It carries a reference number by which we can recognise the file later. You cannot see it, and it cannot be saved away — unlike the provenance data it also survives the re-encoding that happens on upload. Reading it requires the matching tool, so for that you depend on us.

For the labelling to be accurate, Ikrames determines where each post comes from. A carousel of text and layout is fully AI-generated. Put your own photo in it and it becomes a mix — and is labelled “AI modified”, not “AI generated”. That difference is not a formality: declaring your own photo as an AI creation would be just as wrong as a missing label, only in the other direction.

If you want it, Ikrames also appends a short note to your post text — “AI-generated” or “Edited with AI”, whichever applies to the post. In your brand language, without a hashtag, and placed so it does not fall victim to the platform's character limit. By default your text stays untouched — the visible icon does the labelling.

You can check this yourself. Upload a file from Ikrames to the official verification tool — it shows you the same data a platform would read.

Verify Content Credentials

When verifying, you will see that our certificate is not yet on the official trust list — the provenance data is readable, the issuer is us. We are moving to a recognised certificate; what the data says will not change.

One limitation we will not hide: many platforms re-encode files on upload, and the Content Credentials can be lost in the process — the watermark usually survives it, but „usually" is not a guarantee. That is not a fault on your side or ours — it is the reason the visible icon still matters.

What you are responsible for

The moment you publish, you are the deployer under the regulation. If content depicts people, places or events in a way that could be taken for real — the person does not have to exist — it must be visibly labelled as AI-generated when posted (Art. 50(4)). Ikrames can do that for you — the responsibility that it happens stays yours.

Visible means: recognisable to your audience, at the latest on first contact with the content. An icon in the image, a note in the caption, or the platform's own labelling feature — one of them has to apply.

And one more thing that is easily missed: the duty attaches to WHERE content arrives — not to which market you select. Platform algorithms distribute your posts to people you never aimed at. Anyone posting outside the EU who is seen there anyway can be covered by the duty.

This is also set out contractually in Section 7 of our GTC: visible labelling when you publish is your responsibility, and the machine-readable marking must not be intentionally removed.

When it does not apply

• Purely private and non-professional. Posting only for yourself or friends does not make you a deployer under the regulation.
• Artistic, creative, satirical or fictional. A lighter form is enough here — a credit or a note that does not disturb the work.
• Recognisably not real. Only what is obviously an illustration — stylised, drawn, visibly artificial — needs no deepfake label. Important: no existing person has to be shown. If a generated person looks like someone who could exist, the content counts as a deepfake and must be labelled.

When in doubt, label it. It costs you one line and takes the question off the table.

What you can configure

Available

• The target market per brand. If you serve EU markets, Ikrames labels automatically from then on — and explains once what the duty means. If you select no EU market, you get a shorter note instead: that algorithms may carry your posts into the EU anyway, and the duty can then apply. Either way we record what you were informed about.
• The visible icon in images and video. For AI-generated posts, Ikrames places one of the official EU icons in the corner when publishing — burned in, so it survives resharing and downloading. In carousels every slide gets one, because slides are shared individually.
• How the icon looks: corner, size, which of the three variants, a short text suffix of your own and — for video — whether it stays for the whole clip or only appears at the start. We do not recommend less than two seconds; it can be set from half a second, and we ask once if you do. A preview shows you the placement.
• A switch per post. Normally every post follows your brand setting — on the post you can see what that currently means. You can deviate for individual posts; if you turn labelling off in an EU market, we ask once and record your decision.
• Your original stays untouched. Ikrames writes the labelled version as a separate file next to it and publishes that one — you keep the unlabelled one.
• The official EU icons for your own use. There are three: a basic icon, one for fully AI-generated and one for partially AI-modified content. The Commission provides them freely — SVG and PNG, four colour variants each, without attribution and at no cost. Using them is voluntary; the labelling duty itself is not.
• The platforms' own AI switches. TikTok, Instagram, X and YouTube have dedicated fields for declaring a post as AI-generated on upload. Ikrames sets them automatically — based on how the post was created, and independently of whether you turned visible labelling on. The reason: the platforms require this declaration worldwide, not only in EU markets. On Instagram and X the platform then shows its own notice on the post — so alongside our icon there will be one of theirs. What the platform does with the declaration is up to it; we transmit it.

Download the EU icons

In progress

• Labelling for audio-only posts. The regulation also covers AI-generated speech; today Ikrames marks audio machine-readably, but there is no audible cue.
• A certificate from a recognised authority, so verification tools stop showing our marking as “not verifiable” (see above).

These points are not available yet. We name them here so you know what to wait for — and what you need to watch yourself until then.

Sources

• Article 50 of the AI Act (official text)
• Code of Practice on Transparent AI-Generated Content (Commission)
• The three EU labelling icons (download)

This page explains how we implement labelling. It is not legal advice — whether and how you must label in a given case depends on your content and your market.