Concerns Over AI Postcards at Junior Eurovision
by James Stephenson
Additional reporting by Gabe Milne, Sem-Anne van Dijk, Daniel Stridh and GJ Kooijman
The use of generative AI in the postcards of this year’s Junior Eurovision Song Contest in Madrid has come under scrutiny. At the contest, which was held in Spain on 16 November 2024, the show’s postcards intercut real footage of the competing artists, all between the ages of 9 and 14, with AI-generated sections of the performers.
However, questions have emerged over why AI was used in this year’s postcards, who was responsible for creating them, and whether the data captured of the child performers was rendered locally or via the cloud.
If the postcards were rendered locally, the data captured would be stored on a local machine controlled by the creators. But if the postcards were rendered using a cloud-based AI model, the data used to create the AI clips could then be used as training for the model, giving any of its users the ability to create images trained on children.
This would mean that data captured from minors would be openly accessible, raising ethical questions about how the data was handled and the safeguarding of the performers. Additionally, there is no known evidence that the performers, nor their families, had given explicit consent for their image or likenesses to be used to create AI imagery.
EBU Explanation Cast Into Doubt
As of Monday 18 November,, the European Broadcasting Union (EBU), the organisers of Junior Eurovision, say that the postcards were rendered locally:
“To create the AI imagery seen in the postcard’s for this year’s event, pictures of the artists were shared by delegations and were then stored locally on the machines of the company commissioned to make them and never in a cloud.”
They further elaborated in a statement sent to Eurovision journalist Matthew Joyce that the company, who is not named, used AI softwares called Runway and Stable Diffusion to create the postcards.
“All the data developed to train the AI models was done locally using software called Stable Diffusion.
Animations were then created on local machines in software called Runway by using the avatars created in Stable Diffusion, not the original images of the artists.”
On the surface, the EBU’s statement appears to be a credible explanation of the production process. However, technical experts and sources close to the contest’s production dispute the EBU’s claims.
To understand why, we need to introduce you to a different AI software: Luma. Luma is a company which uses a cloud-based AI model that does not have the functionality to run on local machines. That means any imagery created in Luma is created in the cloud, necessitating the uploading of data.
And, although the EBU’s statement does not name Luma, this shot from the broadcast postcard of The Netherlands’ act, Stay Tuned, clearly shows Luma’s watermark in the top right corner. This was first discovered and reported by Daniel Stridh.
The Luma watermark does not appear in any other part of Junior Eurovision’s production. However, this watermark would only be displayed if using the free version of Luma’s “Dream Machine” model.
Luma’s terms and conditions, available on their website, make it clear that anybody who uses Luma gives them permission to use any data inputted into its software to "create, test, improve, train, or otherwise develop the artificial intelligence or machine learning models".
That means, if Luma’s AI model was used to create any of the postcards at Junior Eurovision 2024, any images that the delegations provided of the performers would have been inputted into Luma’s AI model, making that data accessible to anybody using the software.
Along with the evidence that the Junior Eurovision postcards utilised Luma, sources say that the EBU’s explanation of how the postcard creators used Runway is not technically possible.
This is because the Runway software, which the EBU says has been used locally, cannot be run locally. That’s at least according to the software’s developers, who clarify in the security and privacy standards on their own website that Runway is “a cloud-based browser service”.
In this guide, Runway specifies that the media uploaded to its system is secure and cannot be accessed or extracted by third parties:
“If you’re editing videos for your own clients through a personally-owned Standard, Pro, or Unlimited Plan, our SOC 2 compliance guarantees that all uploaded media is not accessible or extractable by unauthorized team members and/or third parties.
For those on an Enterprise Plan, this compliance still applies, but any additional requested, discussed, or amended plan-based security details (especially involving media processed on the Runway platform for said clients) are specifically outlined in your original enterprise contract.”
But while this guide makes it clear that the assets you create using Runway are secure, it doesn’t provide any information about additional security for the data you upload into the cloud-based Runway software to create the assets themselves.
Furthermore, according to comments from users on Reddit, the functionality to use Runway on a local machine had been removed over a year ago by the software’s developers.
“In the early versions of Runway you could use your own CPU or GPU for some of the models. They removed this functionality.”
The EBU’s statement says that the postcard creators used Runway’s software on local machines, using AI avatars created in Stable Diffusion. As a result, the creators could have uploaded the avatars into Runway, and not the images of the child performers themselves. But if that were correct, it would still mean that assets based on the data of minors was uploaded into a generative AI software
And, as evidenced above, at least one shot broadcast at Junior Eurovision is proven to have been created using the entirely cloud-based Luma AI software. The EBU’s statement, then, raises significant questions as to how the postcards could have been created while protecting and safeguarding the competing artists’ data.
Who Was in Charge of The Postcards?
In their statement, the EBU claimed that a third party company was responsible for the creation and production of the AI in the postcards. The company in question is Creative Works London, a UK-based creative studio.
The company’s co-founder and the Executive Creative Director of the postcards, Dan Potter, confirmed that his team had worked on the project in a LinkedIn post. He described their work on the project like this:
“We developed a content package solution that seamlessly wove together film, music, motion graphics, and AI to bring out the concept of 'Let's Bloom' and capture a snapshot of aspiring music performers' character, creativity, and craft.”
The Executive Producer of Junior Eurovision 2024, Ana María Bordas, confirmed that Creative Works London was chosen for the project because they wanted “a company specialising in this type of creation with a lot of experience behind it”.
Creative Works London are credited with creating imagery, through the use of generative AI, for some of the world’s biggest artists, including Guns n’ Roses. In an interview with Decrypt, Potter spoke about his work with the band on the music video for their song “The General”, outlining how CWL integrated Stable Diffusion into their workflow on the project:
Stable Diffusion, unlike Runway and Luma, can be run on local machines. This means that the data used to create the animations in Stable Diffusion would not need to be uploaded to any cloud server.
However, in his post about CWL’s work on Junior Eurovision, Potter does not specify the workflow they used to create the Junior Eurovision postcards. Additionally, there is no evidence that Creative Works London has ever utilised Runway’s AI software in any previous project.
It remains unclear as to how Creative Works London came to be involved in this year’s Junior Eurovision Song Contest. However, the company did have an existing relationship with a member of its production team.
In this post, Potter speaks about another project Creative Works London was part of - the visuals for a tour by German artist Ayliva. It confirms that their work on Ayliva’s tour was done in collaboration with Marvin Dietmann - the Artistic Director of this year’s Junior Eurovision Song Contest.
Dietmann has long been associated with both Junior Eurovision and the main Eurovision Song Contest. An article by fan media website Eurovision Fun claims that Dietmann staged 10 performances at the Eurovision Song Contest 2023 in Liverpool, United Kingdom, showcasing his influence on the production of the show.
As this year’s Artistic Director, Dietmann played a major role in deciding the direction of the show’s production. Sources with knowledge of Junior Eurovision’s production have said that Dietmann had final approval on many elements of the show, demonstrating his influence on the look and feel of the contest overall.
Additionally, in her interview with verTele!, Executive Producer Bordas claimed that the initial idea for the AI postcards came from the Artistic Direction team. As leader of that team, and based on his previous relationship with Creative Works London, it’s clear Dietmann was influential in the creation of the AI postcards.
Postcards Show Clear Errors and Cut Corners
When the postcards for Junior Eurovision were first shown publicly, during the dress rehearsal for the contest on Friday 15 November, they faced negative reaction on social media.
Many users on the social media platform X criticised the use of AI in the postcards, with some suggesting that the use of generative AI in postcards involving children was particularly concerning to them.
However, closer inspection of the postcards used at Junior Eurovision this year reveals several technical errors and shortcuts taken by their creators.
Firstly, it became apparent during the jury show that the postcards and the title cards, showing the key information about the songs, were separate files. This was made clear by an abrupt cut in audio when switching between them on the broadcast, although this issue was resolved in time for the final.
Some of the postcards also contain visual errors or strange imagery. For example, the postcard for the artist who won the contest, Georgia’s Andria Pukharadze, shows him having four fingers in one shot, while Ireland’s entrant’s hair colour is also wrong.
(Andria Pukharadze - Georgia)
(Enya Cox Dempsey - Ireland)
Another example shows that the AI software used in the production created “artefacts” in various areas. These artefacts are created when the machine learning model makes errors in how it represents imagery. For example, the title card of the Albania postcard shows her walking off frame, while the image of the artist appears melted into a plinth behind her.
(Nikol Çabeli - Albania)
Some errors were caught by the postcards’ producers between the jury show and the Grand Final the following day. But the way those errors were hidden often relied on shortcuts.
In this example, broadcast in the jury show, the male member of The Netherlands’ representatives Stay Tuned starts to turn into a blonde girl during the title card. For the Grand Final, this was fixed by shortening the AI section of the title card before they transition into real-life footage. However, the real-life footage has clearly been superimposed over the faulty AI imagery, which is still present behind.
(Stay Tuned - The Netherlands (Dress Rehearsal 2))
(Stay Tuned - The Netherlands (Grand Final))
That means Creative Works London and the Junior Eurovision team were still making edits to the postcards with hours to go until the live show. However, this also means that many other errors, including the ones listed above, were not fixed.
There is also evidence of Creative Works London reusing assets from previous projects at Junior Eurovision. In the full version of the LinkedIn post we showed above, Dan Potter goes on to describe a set piece CWL created for Ayliva’s tour, featuring imagery from the live performance:
This same concept was re-used almost entirely for the opening act of Junior Eurovision 2024, which Potter has confirmed Creative Works London were commissioned to create. During the opening, previous Junior Eurovision winner Zoe Clauzure emerges from a similar structure.
Furthermore, sources close to the production say that the competing delegations at the contest didn’t see the finished postcards until the last possible moment. The first time delegations saw the postcards was when they were used during the first dress rehearsal for the contest, which took place earlier on Friday.
Sources have claimed that multiple delegations immediately made complaints about the postcards once they viewed them, as well as other elements of the show’s production. However, by this point in the production, it was too late for any significant changes to be made.
Delegation sources are also reporting that there were several complaints from delegations because the changes requested after the stand-in rehearsals weren't carried over to the first rehearsals.
AI Also Apparent in Staging Backgrounds
The postcards weren’t the only element of the staging that utilised generative AI, though. Several of the performances at Junior Eurovision also integrated AI into their staging, with many countries having AI-generated assets on the large Unlike the postcards, the individual staging concepts were not worked on by Creative Works London. Instead, each competing country had a choice to make: they could create their own staging or hire a third party to make it for them, or commission their staging from Junior Eurovision’s host broadcaster, RTVE.
At Eurovision, staging a song comes with a price tag. That means the national broadcasters with higher budgets can produce more impressive stagings, while countries with less money available have fewer options.
Italy, whose broadcaster Rai is a member of the “Big Five” at Eurovision, was able to commission Gio Forma, who created the staging for Måneskin’s Eurovision winner “Zitti e Buoni” to create theirs. Many countries, however, opted to commission RTVE to create their backgrounds to cut costs.
Using AI in Eurovision staging can also reduce the cost of the stagings themselves, meaning a lower price tag for less well-off broadcasters. Additionally, the option to use generative AI in backgrounds can create a more even playing field, giving those broadcasters the opportunity to create spectacular visuals without a high budget.
However, many countries found that their staging backgrounds, much like the postcards, contained errors as well. For example, Estonia’s entry “Tanavad” used a background of a New York-like cityscape, but the AI model used to create this included a building that closely resembles one of the towers of the former World Trade Center.
(Estonia - Grand Final)
Even more notable was Poland’s performance, in which singer Dominik Arim was digitally aged from a child to an old man on the LED screen in the space of a few seconds as he sang. This resulted in a strange, surreal visual that dominated the staging.
Generative AI is nothing new for Eurovision, or Junior Eurovision. At last year’s JESC in Nice, France, Armenian entry Yan Girls utilised AI in the music video for their song “Do It My Way”, while the aforementioned RTVE also created AI backgrounds for their own Eurovision entry “Zorra” this year.
But, much like the postcards, the backgrounds were also criticised on social media and showed errors. And, along with other staging elements, the changes requested after the stand-in rehearsals weren't carried over to the first rehearsals.
AI Raises Ethical Concerns for Eurovision
The Junior Eurovision Song Contest is not the first, and will not be the last entertainment programme to attempt to integrate AI into its production.
Since it emerged, generative AI has revolutionised the way digital art can be created. Thanks to accessible, free-to-use tools alongside more advanced AI models, both individual people and creative businesses can use AI to create groundbreaking visuals quickly and cheaply, in some cases replacing the need for traditional artists and graphic designers.
However, there are ethical concerns over how these AI models create this imagery. Most AI models use machine learning, with developers scraping millions of images and videos across the internet to train the AI to create believable images.
And, while AI is expected to shape many aspects of our lives in the future, the technology is little understood by most people. There are hundreds of AI models out there, each working in a different way and integrating differently with other technology.
As a result, it can be easy for companies not to truly appreciate the impact of AI. In the face of this complex web of information, even large multi-national organisations like the EBU may not have the knowledge required to understand the processes and technicalities of the tools they are utilising.
One of the biggest impacts of AI is what the images it creates can represent. Generative AI has been criticised for “beautifying” its subjects, making them appear more in line with traditional Western beauty standards. In the postcards for Junior Eurovision, we found multiple examples of where artists, particularly girls, had been made to appear skinnier and more blonde in the AI footage.
(Enya Cox Dempsey - Ireland)
(Maria Pissarides - Cyprus)
(Stay Tuned - The Netherlands)
The postcards also feature multiple instances of the artists’ ethnicities being changed. Artists such as Italy’s Simone Grande, who has an olive skin complexion, appear in their AI-generated selves as having whiter skin than in real life. On the other hand, North Macedonian artist Aleksej is generated as having darker features than their real-life self in a shot from their postcard:
(Simone Grande - Italy)
(Aleksej (of Ana & Aleksej) - North Macedonia)
There are further instances where young female performers have been unnecessarily altered in their physical appearance to come across more mature in their postcards. Cyprus’s artist, Maria Pissarides, is generated by AI in a shorter item of clothing in the shot below, while the title card of Ireland’s Enya Cox Dempsey emphasises particular features of her body so that it appears more mature.
(Maria Pissarides - Cyprus)
(Enya Cox Dempsey - Ireland)
All of these examples highlight worries that AI can modify its subjects to make them appear “better” than their true selves. In a world where there is increasing pressure and mental health issues derived from body image problems, especially in young people, these postcards raise questions over the example they set for Junior Eurovision’s audience.
More pressingly, these postcards are also using the images of minors, who the organisers of Junior Eurovision have a duty of care towards. As it stands, the European Broadcasting Union, the Junior Eurovision host broadcaster RTVE of Spain, or Creative Works London have not provided any evidence that specific consent was given for these images to be created.
While the statement sent by the EBU on Monday night says that “pictures of the artists were shared by delegations”, it doesn’t specify whether they knew what the pictures were being used for. The EBU has not responded to an email we sent them on Friday before the live show occurred asking, among other questions, whether explicit consent had been given for images of the child performers to be fed into an AI model.
These concerns are why how the AI footage was rendered is important. If the images were rendered using a cloud-based model, without prior consent from artists’ legal guardians and delegations for them to be used specifically for that purpose, the postcard producers and the EBU would face strong questions on the ethics and legality of their actions.
Finally, the decision to integrate AI into the postcards at Junior Eurovision goes against previous comments the EBU has made on AI.
In 2023, Jean-Philip de Tender, Deputy Director General of the European Broadcasting Union, stated in an interview at the Edinburgh TV Festival that the EBU was debating banning generative AI from Eurovision altogether.
During the interview, he said he was “reflecting on how we need this [AI] in the rulebook, that the creativity should come from humans and not from machines.”
The fallout from the postcards at this year’s Junior Eurovision Song Contest has demonstrated the questions that arise when AI is used to augment or enhance entertainment. Now, the EBU must grapple not only with mounting questions about how AI has been used in this instance, but whether they choose to use this technology in the coming years, and how they do so ethically.
