Video EffectsMandatory Labeling of AI-Generated Content: What Changes for Social Media Publishers
Team at Video EffectsWritten with AI assistance18 min readIn summary
- The Article 50 of the European Artificial Intelligence Regulation (AI Act) applies from August 2, 2026 and requires notification when content has been generated or manipulated with AI (European Commission).
- There are two distinct obligations: the machine-readable mark, which is provided by the tool's supplier, and the visible label, which you put when publishing. Complying with one does not exempt you from the other (Official Commission FAQ).
- The transitional period of four months for systems that were already on the market before August 2 to incorporate the technical mark of Article 50(2) ends on December 2, 2026. Anything launched after August 2 must have complied from day one.
- If you animate a real photo —your store, your product, your face— and the result can pass as authentic, you are facing a deepfake in the sense of Article 3.60 and you need to label it. Minor color adjustments or a crop do not count.
- Fines for non-compliance with Article 50 can reach 15 million euros or 3% of global turnover, with explicit proportionality for SMEs (Commission FAQ). In Spain, the bill that established the national sanctioning regime expired on October 6, 2026.

For years, posting on social media was a matter of judgment: what to tell, when, and with which photo. Since the summer of 2026, there is an additional layer, and it is legal: if the content has been created or edited by an AI and can seem real, it must be stated. This affects the bakery that animates a photo of its display for a reel, the agency that sets up the campaign, and the community manager who schedules the post. At AIsToHire, we keep a close eye on this because the tool we maintain for this work, Video Effects, does exactly that: it turns your photo into a video with effects; and the video that comes out of it is, almost always, content that needs to be labeled when published. If you want to see the tool before continuing, here’s how to convert a photo into a video with AI effects. The rest of this article is about the regulation, not the app: what Article 50 says in plain Spanish, who it obliges, what the machine-readable mark is, and what exactly happens when all you've done is give movement to a photo that is yours.
What Article 50 of the AI Act says and from when it applies
Article 50 of Regulation (EU) 2024/1689 requires transparency in four situations: when someone talks to an AI, when a system generates synthetic content, when emotions are recognized or biometric data is categorized, and when a deepfake or AI-generated text on matters of public interest is published. It applies from August 2, 2026, according to the guidelines on transparency obligations from the European Commission.
Two important clarifications that are often confused:
- The Digital Omnibus did not delay Article 50. The simplification package approved in 2026 postponed the obligations for high-risk systems in Annex III to December 2027 and those for AI embedded in products to August 2028, but left the transparency of Article 50 intact, which remains in force since August 2026 (analysis of the postponement).
- There is no retroactive labeling. Content generated before August 2, 2026, does not need to be labeled retroactively, according to the FAQ of Article 50 published by AESIA.
The timeline that affects you
| Date | What happens |
|---|---|
| June 10, 2026 | The Commission publishes the Code of Good Practices on Transparency of AI-Generated Content and the set of official EU icons |
| August 2, 2026 | Article 50 is fully applicable: chatbot notification, deepfake labeling, technical mark on new systems |
| December 2, 2026 | End of the four-month transitional period for the machine-readable mark of Article 50(2) in systems that were already on the market before August 2 |
| December 2, 2027 | The high-risk obligations of Annex III, postponed by the Digital Omnibus, come into effect |
Sources: European Commission, Official FAQ of Article 50.
Who it Obligates: Provider, Deployment Responsible, and Personal Use
Article 50 divides the work between two figures. The provider is the one who develops the AI system and puts it on the European market under their name or brand: they are the ones who must embed the machine-readable mark in everything their system generates. The deployment responsible (what the regulation calls deployer) is the one who uses that system under their own authority in a professional context: they are the ones who must put the visible label when publishing. If you are a small business, an agency, or a creator who makes a living from this, you are responsible for the deployment.

The boundary with private use is in economic benefit. The official FAQ from the Commission states it clearly: individuals using AI systems for non-professional personal activities are excluded, but if they obtain economic benefit regularly or act as freelancers or businesses, they become responsible for deployment and are subject to the obligations.
| Profile | Responsible for Deployment? | What they have to do |
|---|---|---|
| Small business publishing on their business Instagram | Yes | Visible label on each piece that requires it |
| In-house community manager | Not responsible separately: the company is | Apply their company's labeling policy |
| Freelancer managing client accounts | Works under client control; does not become separately responsible | Agree in writing who labels |
| Creator monetizing their account | Yes, if there is regular economic benefit | Visible label |
| Individual animating a photo for the family group | No | Nothing |
A nuance of scope that catches many people off guard: for deepfake labeling, it is enough to publish on open internet, without the content needing to be specifically directed at the European market (Faegre Drinker).
What is a Machine-Readable Mark and Why it Does Not Replace the Label
A machine-readable mark is a signal embedded in the file itself that allows a program to detect that the content is artificial: an invisible watermark in the pixels, signed provenance metadata like the Content Credentials from C2PA, or a digital fingerprint of the file. It is not visible to the naked eye: it is designed for detection by platforms, not by people. Article 50(2) requires that this mark be effective, interoperable, robust, and reliable, and that it withstands common alterations: cropping, compression, resolution changes, and format conversions. The provider must also offer free detection tools with trust scoring.

Here is the most expensive mistake a small business can make: thinking that, since the tool already marks the file, the job is done. It is not. The official FAQ for Article 50 is explicit: public disclosure must be understandable and perceivable by individuals —visible or audible labels— without the need for technical tools, and the machine-readable mark by itself is not enough. Those responsible for deployment cannot hide behind the mark inserted by the provider.
| Machine-Readable Mark (art. 50.2) | Visible Label (art. 50.4) | |
|---|---|---|
| Who puts it | The provider of the AI tool | Whoever publishes the content |
| Who perceives it | Programs and platforms | People |
| How it is done | Invisible watermark, C2PA metadata, digital fingerprint | Icon, label, notice in the video or next to the publication |
| Deadline | August 2, 2026 (new systems) · December 2, 2026 (systems already on the market) | August 2, 2026 |
| Does it replace the other? | No | No |
The Code of Good Practices on Transparency of AI-Generated Content, published on June 10, 2026 and signed by about 190 companies and organizations by the end of July 2026, is voluntary, but the obligations of Article 50 are not: those who do not adhere must demonstrate compliance through suitable alternative means. For content published online, the Code requires at least two layers of marking —metadata plus watermark— given the current state of the art (Orrick).
What if I just animate a photo of mine for a reel?
It's the question that gets asked the most, and the uncomfortable answer is: yes, it's normal to have to label it. That the photo is yours, that the person in the image is you, and that there is no intention to deceive does not change anything. Deepfake is defined in Article 3.60 of the AI Act as image, audio, or video content generated or manipulated with AI that resembles existing people, objects, places, or events and that, at first glance, could appear authentic or truthful. Animating a real photo of your establishment or your product fits perfectly.

Three cumulative criteria, according to the official FAQ: high resemblance to existing people, objects, places, or facts; that what is represented exists or could plausibly exist; and that it appears falsely authentic or truthful. And two warnings from the Commission's guidelines that are worth internalizing:
- No intention to deceive is required. The obligation to label applies equally (Greenberg Traurig).
- The criterion is how the public perceives it, including vulnerable groups, not what you intended when publishing it.
Ownership of the original photo is not an exemption. What triggers the obligation is that the result could pass as a real video.
What has a softer rule is the manifestly artistic, creative, satirical, or fictional work: there, Article 50(4) reduces the obligation to disclose the existence of manipulated content "in a manner that does not hinder the display or enjoyment of the work." Note: it is a reduced obligation, not an exemption. A commercial reel of your business is not a satirical work.
What counts as "standard editing" and is excluded
Article 50(2) exempts AI systems that perform a function of assistance in standard editing and that do not substantially alter the input data or its semantics. It is the escape valve for normal retouching, and it is narrower than it seems: as soon as the change modifies the meaning of the image, it falls out of the exemption.
The Commission's guidelines provide specific examples on each side of the line (compiled by Orrick):
| Excluded (standard editing) | Included (manipulation) |
|---|---|
| Spelling and grammar correction | AI-generated summaries and paraphrasing |
| Minor cropping and video rescaling | Removal or replacement of objects in the image |
| Normal color and contrast adjustments | Replacement or modification of faces |
| Noise reduction, red-eye correction | Composed images from multiple sources |
| Format conversion, minor polishing | Realistic voice synthesis |
| — | Extreme lightening or darkening that changes the meaning of the message |
To put it in professional terms: passing your photo through a color filter is still retouching. Turning that photo into a video with movement is generating new content, and that is already included.
For text on matters of public interest — politics, health, justice, security — the exit is different: there is no need to label if it has undergone human review and editorial control with identifiable editorial responsibility, with an identifiable person or organization and accessible contact details. The guidelines warn that spell or grammar check is not sufficient to consider that there was human review (Greenberg Traurig).
Where and how the label must be displayed
The label must be at the latest at the moment of the first exposure, in a clear and distinguishable form, and perceivable without special tools. Article 50(5) sets that moment; the guidelines specify the placement according to the format, and the detail matters: a notice buried in fine print or hidden three clicks down is not acceptable.
According to the placement outlined in the guidelines (Orrick) and the AESIA FAQ:
| Format | Where the disclosure goes |
|---|---|
| Video | At the beginning, after interruptions and at regular intervals, embedded or with equivalent overlay |
| Image | Clearly perceptible icon in the first exposure, with no overlays covering it |
| Published text | Above or at the beginning of the text, alongside the headline, or in the colophon |
| Audio | Audible notice, perceivable without tools |
Two requirements that are often forgotten: the icon must remain visible when the content is reshared or downloaded, and it cannot be hidden by another layer of the interface.
The Commission published on June 10, 2026 a set of three official EU icons — one base generic AI indication, one for AI-generated content, and one for AI-modified content — available in SVG and PNG, free of charge and without attribution, in Annex 1 of the Code of Good Practices. Using them is optional; labeling is not (European Commission).
What platforms are already doing (and why it doesn't cover you)
TikTok, Instagram, and YouTube have had their own synthetic content labeling systems since before the AI Act, and they largely work by reading the technical mark of the file. That helps, but the obligation of Article 50 is yours, not the platform's: if the automatic system detects nothing and you declare nothing, it is you who is in breach.
- TikTok requires declaring AI-generated content that shows scenes or people with a realistic appearance, offers its own self-labeling tool, and applies automatic labels by reading the Content Credentials of C2PA (official TikTok policy on AI-generated content, agreement with C2PA).
- YouTube mandates declaring altered or synthetic content that appears realistic upon upload, in the attributes section, and then shows the label to the viewer; declaring it does not limit the audience or monetization (YouTube help on altered or synthetic content).
- Instagram and Facebook use the label "AI info" and require self-declaration in photorealistic video or realistic sound audio; when the system detects that the content was only modified with AI, the label moves to the post menu, making it less visible (guide on how Meta decides to show the label).
That last point is the one to remember: a label that the platform places in a submenu hardly meets the standard of "clear and distinguishable at first exposure" from Article 50(5). If the format allows, add the disclosure in the video itself or in the copy, in addition to activating the platform's switch.
Sanctions and who is monitoring in Spain right now
Non-compliance with Article 50 is sanctioned with fines of up to 15 million euros or 3% of the total global turnover of the previous financial year, whichever is greater, and the official FAQ states that proportionality will be considered for SMEs and small mid-cap companies. No one expects an eight-figure fine for a bakery, but the risk is not zero.
In Spain, the situation has a current detail worth knowing: the Organic Law Project for the proper use and governance of artificial intelligence, which designated authorities and set the national sanctioning regime, was presented on May 28, 2026, published in the Official Bulletin of the Cortes on June 12, 2026, and expired on October 6, 2026 with the dissolution of the Cortes, after six extensions of the amendment deadline (processing file, Council of Ministers). The next government will have to present a new text.
That national development being stalled does not suspend anything: the AI Act is a European regulation with direct application and Article 50 has been binding since August 2, 2026. What remains in limbo is the detail of the Spanish sanctioning procedure, not the obligation to label.
Practical checklist for publishing an AI-generated video
If you manage the social media of a business and do not have legal advice by your side, this is the minimum operational:
- Classify the piece. Can it pass as real? If there are existing people, places, or products and the result seems authentic, treat it as deepfake and label it.
- Check that your tool marks the file. Ask the provider if it embeds invisible watermark and provenance metadata, and if it offers a free detector. It is their obligation, but you need to know.
- Make the label visible yourself. Do not delegate to the technical mark or the platform's detector.
- Place it at the beginning of the video and repeat it if the video is interrupted or long.
- Also activate the platform's switch (TikTok, YouTube, Instagram): it is compatible and does not penalize monetization on YouTube.
- Put it in writing. An internal policy of one page: what is labeled, with what text or icon, who reviews it. If you work with freelancers, agree on who labels.
- Keep a record. What tool, what source photo, what date. If someone asks, you don’t want to reconstruct it from memory.
None of this requires a lawyer. It requires routine.
Frequently Asked Questions
Do I have to label a video if the original photo is mine?
Yes, in most cases. Ownership of the photo is not an exemption: what triggers the obligation of Article 50(4) is that the result resembles existing people, objects, or places and could appear authentic. If you animate a photo of your premises and the video could pass for a real recording, label it.
Is the invisible watermark provided by the tool sufficient?
No. The official FAQ from the European Commission explicitly states that the machine-readable mark is not sufficient on its own and that the disclosure must be perceivable by people without technical tools. These are two distinct obligations and fall on different parties: the provider marks the file, the publisher adds the visible label.
What happens on December 2, 2026?
The four-month transitional period for providers whose generative AI systems were already on the market before August 2, 2026, to comply with the machine-readable mark of Article 50(2) expires. It does not affect the obligations of the publisher, which are enforceable from August.
Will I be fined if I forget to label a reel?
Article 50 provides for fines of up to 15 million euros or 3% of global turnover, with proportionality for SMEs. In Spain, additionally, the bill that specified the national sanctioning regime expired on October 6, 2026, so the national procedure is pending; the European obligation, however, applies equally.
If I publish from Spain but my audience is outside the EU, does it apply?
For the labeling of deepfakes, publishing on the internet with global access can trigger the obligation without the need to specifically target the European market, according to the analysis of the final guidelines from the Commission. In practice, if you publish from a Spanish company, label it.
Do I also need to declare a color filter or a crop?
No. Normal adjustments of color and contrast, minor crops, rescaling, noise reduction, or red-eye correction are considered standard editing assistance functions and are excluded. The line is crossed when the change alters the meaning: replacing faces, removing objects, composing images, or turning a photo into a video with movement.
What if the video is clearly humorous or fictional?
Article 50(4) reduces the obligation for manifestly artistic, creative, satirical, or fictional works: it is sufficient to reveal the existence of the generated content in a way that does not hinder the enjoyment of the work. It is a reduced obligation, not an exemption, and commercially branded content hardly fits there.
Who labels if I manage a client's social media?
The obligation falls on the party responsible for the deployment, which is usually the company that owns the account. A freelancer working under the client's control does not become separately responsible, but it is wise to have in writing who adds the label to each post so that it does not fall into a grey area.
Do it right from the photo you already have
Complying with Article 50 does not mean stopping the use of AI: it means using it and stating it. The healthy flow for an SME is the simplest: start with your own photos —your product, your premises, your team— generate the video, and label it when publishing. This instantly avoids half the problems because you are not fabricating people or scenes that do not exist; you are simply giving movement to something of yours, and you declare it.
For that part, there is Video Effects, the tool from AIsToHire designed just for that case: you upload a photo from your mobile, gallery, a file, or the camera, choose one of the 57 available effects —22 for movement and 35 viral— and download an MP4 in 9:16, 4:5, 1:1, or 16:9 ready to publish. Without editing and without writing prompts. You add the label when uploading it to the network, with the official EU icon or a clear label at the beginning of the video. If this is the flow you are looking for, start by creating videos with effects from your own photos.
Turn your own photos into videos with effects and publish them properly labeled. Try Video Effects for free: upload a photo, choose the effect, and download your MP4 ready for social media.
Turn your own photos into videos with effects and publish them properly labeled. Try Video Effects for free: upload a photo, choose the effect, and download your MP4 ready for social media.Keep reading
View all articles →
How to animate a photo with AI and turn it into a video: which effect to choose based on the photo you have

Why Your Car Insurance Increased in 2026: Average Premium of €1,412 and Historic Record

Discussing via WhatsApp or talking in person: what science says about couple conflicts through messages
