“Enough just complaining about politics! Turn your indignation into content that thousands will watch and share.”
The promise prints courses ads that teach you how to create political videos with (AI) and, with them, to profit on social networks.
The classes come with instructions in the logic “copied, coloted, generated” so that the IAS produce the contents. The “viral prompts” have characterizations such as “aging skin, wrinkles and iconic white beard” to the president () and “lightly frowning eyebrows and contracted lips, emphasizing intensity and tension” for the former president ().
Among the tips offered are also the replacement of sensitive words and the use of grok,. This is because the chatbot integrated into the social network (former) generates realistic images of famous, while most other Ias blocks such apparitions.
Tests performed by Sheet They showed that Sora, OpenAi’s video generation model, owner of the, also produces images with faces of public agents.
Sought, OpenAi claimed to have additional barriers for political candidate videos and prohibit the use of campaign tool, “including the generation of personalized or targeted campaign materials to specific demographic groups.”
The company said it allows “valuable uses such as education, satire and political comments” and monitors generations on and off the platform. Famous people who prefer their faces not to be inserted in AI -generated videos should speak directly to OpenAi.
“Public figures who want their photorealist representation not to be generated can contact us,” said a company spokesman.
Xai did not respond to contact requests.
The report participated in two video generation courses with AI. One of them, the “AI School”, charges about $ 300 and gives access to video lessons, prompts and participation in a group of whatsapp from alumni with more than 140 members.
Another course, the “Veo3 Dark”, sold for $ 20, delivers a system that analyzes whether videos can be blocked by platforms such as X, Instagram and Tiktok. The user receives specific alerts about excerpts that can generate suspension or removal of the content.
There are some possible ways to create videos taught in the courses. One of them begins to ask ChatgPT, based on newspaper news, make the scenes descriptions that will be portrayed.
These itineraries are transferred to the IAS that produce the images, and the static scenes saw videos with sound through other tools – a time of video work.
Veo3, Google model that creates audio videos directly from text, reduces the process to a few minutes. There is one however: it works only with generic faces and shows sensitivity to mentions of known figures even when they are not being portrayed.
During a demonstration accompanied by the report, an instructor tried to create a scene in which militants asked for “we want mortadella”, mentioning Lula. The command was denied. The solution was to replace the petista’s name with “octopus” in the satirical play.
OpenAi’s Sora, does not produce audio videos. In its policies, the Sam Altman company tool states that it prohibits the generation of videos that use anyone’s image (including politicians) without consent. The company also says it uses multimodal classifiers to analyze text, images and videos to block misleading electoral content.
No caso do Grok, de Elon Musk, “o ‘system prompt’ [instrução interna que o modelo lê antes de responder] Explicitly says that he is not censored and seeks the ‘maximum truth’, even though he is politically incorrect, “says Pedro Burgos, I.’s consultant and professor at Insper.
Xai, the company responsible for the development of Grok, said in early July that it made adjustments to after the system presented answers with anti -Semitic content and exaltation to Adolf Hitler.
For Burgos, Musk identified demand in the technological community by less restrictive models. “Grok makes the opposite of other companies. They think competitors are too carefully concerned,” he says.
The PT joined the videos with IA in late June to promote the taxation “BBB” (billionaires, Bets and Banks) after Congress barred the increase in IOF (Tax on Financial Operations).
More recently, to defend his narrative in the midst of the surcharge crisis imposed by the US President,, Brazil, the party to boost videos of the type.
Centrão members, but the cheap production made the phenomenon irreversible, according to experts.
“Deepfake is not synonymous with content designed to cause damage. There are many artists who use the resource for satire and even shows. The problem is in deceiving,” says Beatriz Farrugia, professor at FAAP (Armando Álvares Penteado Foundation) and researcher for the Lupa agency’s misinformation.
Brazilian law does not yet provide for specific punishments for deepfakes. According to Patrícia Peck, a lawyer specializing in digital law, production can be framed in crimes against honor, sexual crimes or estelionate, “depending on content and intention”.
The TSE (Superior Electoral Court) defined the regulation of AI use in electoral contexts and the prohibition of use of Deepfakes by parties and candidates in official electoral propaganda. For Peck, “the person responsible for the dissemination of irregular electoral propaganda on the internet, even though he is not a candidate, is a legitimate part to respond to representation.”
Digital content creator Matheus Delalibera, 30, is the owner of AI’s school. He is in favor of signaling when the content is generated by AI, “especially for older people, who cannot differentiate what is not” real.
“I never want to use the course for evil,” says the platform owner.
According to him, the students are mostly men – nor old seeking solutions for business and young people wanting to viral with humor. The contents offered at the AI School range from the creation of satirical deepfakes to the cloning of their own face to entrepreneurs aimed at enhancing sales.
Burgos, from Insper, is skeptical about the malicious use of technology by the campaigns in 2026. He cites the example of (PRTB), reported at the Electoral Public Prosecution Service for the dissemination of a fake medical report against Guilherme Boulos (PSOL) in 2024, when the two ran to the São Paulo City Hall.
The court revoked the mandate of Councilman Rubinho Nunes (União Brasil) for reposting the false document. Marçal is ineligible by other episodes.
“Technology already existed in other elections, and campaigns associated with fraudulent content have a lot to lose,” says Burgos.
A Sheet He tried to contact social networks with those responsible for the Veo3 Dark course, but received no response until the publication of this report.
