Legal · AI Act
The piece must be identified because the AI has generated or altered a relevant part?
KarminaAI StudioAgents · Services · Training
KARMINA × UTOPIQ · PRACTICAL GUIDE
The AI Act is here. What does your brand need to know?
An open guide to decide when to explain the use of AI, when to review rights and when the brand criterion should set the limit.
Article 50Synthetic ContentConsentEditorial judgementHuman review
60 SECOND
Not every use of AI needs a label.
The important question is not whether the team has used an AI tool. It is whether the final piece creates or modifies in a relevant way what the audience sees, hears or can interpret as real.
Transparency also does not replace copyright, consent, data protection or editorial responsibility.
See the practical rule ↓THREE DIFFERENT CAPES
Legality, rights and brand considerations are different questions.
A piece can fulfill one layer and fail in another. The right order is to review them all before posting.
Rights
Do we have permission, license and consent to use the materials, works and people?
Brand
Does the piece fit the voice, values and level of risk that the brand wants to assume?
THE PRACTICAL RULE
Three questions before publishing.
If the answer points to a realistic representation, a relevant alteration or a possible confusion, it is necessary to make the intervention visible and review the case.
- 01
Has the AI only helped to work or created or modified the final content?
- 02
Can you believe that a person, a product, a place or a real event was different?
- 03
Could anyone interpret the piece as a photograph, statement, recording or authentic evidence?
RECOMMENDED INTERN SEMOFOR
From minor retouching to the line you must not cross.
It is a work tool, not an automatic legal qualification. Context can always change the decision.
01
Green
The AI helps to workIdeation, planning, correction, translation or minor retouching with human review. The internal tool does not have to become the protagonist.
02
Yellow
Intervention is VisibleThe AI participates creatively or may generate doubts, even if the piece is fictional. Identifying it is a good practice.
03
Red
It may seem real.AI creates or substantially alters people, voices, products, places or events. It needs a clear and perceptible indication.
04
Black
Tagging doesn't fix itIf rights, consent or validation are lacking, or the piece may be misleading in a harmful way, it should not be published.
PRACTICAL CASES
You do not need to memorise the rules. You need to make sound decisions.
It opens the format you need and contrasts the type of intervention, not just the tool used.
01Texts, posts and articles
Editorial assistance or automatic publishing: human control makes a difference.
+
Texts, posts and articles
Editorial assistance or automatic publishing: human control makes a difference.
Green
Ask for ideas, structure a calendar or correct spelling.
Green
Create a first draft and rewrite it with effective editorial control.
Yellow
Publish almost intact a commercial text generated with AI.
Red
Publish information of public interest without actual editorial review.
Black
Invent an appointment, some data or some statements.
02Photographs and images
A technical touch is not the same as altering what supposedly existed.
+
Photographs and images
A technical touch is not the same as altering what supposedly existed.
Green
Trim, resize, adjust light or eliminate a minor imperfection.
Yellow
Expand a neutral background or replace a sky without changing the meaning.
Red
Add people, products, or non-existent audience in a realistic scene.
Red
Simulate a local or product not yet existing as if it were a photograph.
Black
Using a person’s identity without permission or attributing false behavior to them.
03Videos and reels
Manipulation matters when it modifies what the public believes happened.
+
Videos and reels
Manipulation matters when it modifies what the public believes happened.
Green
Cut plans, stabilize, correct color, or create revised subtitles.
Yellow
Remove an element from the background if it does not change the documentary context.
Red
Get someone to say words they haven’t spoken.
Red
Create a realistic avatar or fictional plans of a real place.
Black
Present a fictitious testimony, an incident or a fake news as real.
04Voices, music and podcasts
Consent is necessary, but it does not eliminate the duty of explaining what the public hears.
+
Voices, music and podcasts
Consent is necessary, but it does not eliminate the duty of explaining what the public hears.
Green
Clean up audio, transcribe or improve a recording without changing words.
Yellow
Create a fictional corporate voice or an assisted musical piece.
Red
Clone a voice with permission when it may seem like an authentic recording.
Red
Do lip doubling or lip syncing by keeping the person's voice apparent.
Black
Clone a voice without permission or simulate sensitive statements.
HOW TO LABEL
Clear, noticeable and there from the first interaction.
The warning must be understood, visible or audible according to the format, arrive on time and be accessible. It is not necessary to hide it behind “more information” or to rely only on metadata.
“Image generated with artificial intelligence.”
“Real image substantially modified with AI.”
“Visual recreation generated with AI. It's not a real picture.”
“Dramatized video with images generated with AI.”
“Synthetic voice generated with AI.”
“Virtual Avatar. He's not a real person.”
Better to make it clear. “Created with AI” explains more than a hashtag, a little-known icon, or an ambiguous formula like “Mirrored with technology.”
BRAND CRITERIA
The law sets a minimum. Your brand needs its own judgement.
An AI style book turns an abstract discussion into repeatable decisions: which formats support AI, which is never generated, how sensitive pieces are tagged and who approves.
We do not identify the AI that only helps us work. We identify it when it creates or modifies in a relevant way what the audience can interpret as real.Build the base of ↗ criteria and permissions
BEFORE YOU PUBLICATE
The right process leaves a record too.
The decision should not be scattered in messages or depend on who is available that day.
The team registers
- 01
What tools have been used and for what purpose.
- 02
What part of the content has been generated or modified.
- 03
Whether it represents real people, products, places or facts.
- 04
If there is sufficient consent and licenses.
- 05
What label was applied and where it appears.
- 06
Who has reviewed the data and approved the final piece.
CHECKLIST
Before you post, ask yourself...
- Does the play include content generated or altered with AI?
- Is a real person identifiable and do we have their consent?
- Is the base asset own, bank or third party, and the license is reviewed?
- Play a play, a voice, or a recognizable character?
- Have we applied the agreed branding criteria for this format?
- If transparency is required, is the warning clear, visible, accessible and appears in time?
DOCUMENT A DAY A DAY
Transparency is a decision about how you work.
Karmina and Utopiq help teams turn general obligations into a style book, an approval circuit, adapted training and a clear way to record decisions.
KarminaUTOPIQ
OFFICIAL FUNDS
This guide is indicative and does not replace legal advice. In case of a sensitive case, it is necessary to review the role of the organization, the system used and the context of publication.