Parties and their consultants have already devised inventive ways to use deepfakes. How will they be deployed in the coming election—and who will create this content? Splainer investigates.
Written by: Samriddhi Sakunia is an Independent Journalist in Delhi. She covers hate crime, tech, and human interest stories. Her work has appeared in Caravan, Rest Of World, VICE, Al Jazeera and other digital websites. You can find her on @samriddhi0809 , write to her @sakuniasamriddhi@gmail.com
On February 20, the official page of the Indian National Congress posted a meme that showed Prime Minister Narendra Modi singing a famous Hindi song—titled ‘Chor’ by Justh. The lyrics were modified craftily to indict Modi of giving ports, airports, power plants, and coal mines to the ‘chor’—as in, Gautam Adabi. The effect—as you can see-is uncanny:
The parody video soon gained 2.1 million views, 100,000 likes, and 3000+ comments on Instagram alone. “Don’t know about 2024 polls, but you guys have won the meme contest,” said one user. Unwilling to concede the political game, the BJP uploaded its video—with Rahul Gandhi’s face digitally morphed onto that of Rashtriya Janata Dal (RJD) leader and former Bihar deputy CM Tejashwi Yadav. A clip of Yadav criticising Chief Minister Nitish Kumar for defecting to the BJP camp now showed Gandhi scolding Chief Minister Mamata Banerjee for going it alone in West Bengal. As you can see, the party’s AI skills left a lot to be desired:
Their technical merits aside, the two videos are a harbinger of the AI-generated tsunami to come. The technology will be used to turn leaders into multilingual orators, send personalised messages to voters and party workers, mock opposition candidates, create “ghost videos” of past leaders and more. For example, this deepfake message from the PM in support of Narayan Rane in Maharashtra:
While the focus has been on the demand for such content, very little has been written on the suppliers of such material. We went in search of the creators of these videos. Are these slick marketing companies or a network of coders working from their basements?
Enter, the Indian Deepfaker
Developers we spoke to say the videos of Gandhi and Modi were likely made using open source or cheap software. They were not the work of a professional. The clips are also a mark of how the industry has become democratized—accessible to anyone with a laptop. Divyendra Singh Jadoun says:
Just a couple of years ago, producing deep fake content cost us a lot of manpower and data. To produce 60-seconds of content, we would take 12-15 days. But today, anybody can create a deep fake video or swap faces with simple tools… and it would take them less than 3 minutes.
Jadoun is a 32-year-old former student politician whose company The Indian Deepfaker creates "hyper-realistic content" production companies, OTT platforms, Indian advertising agencies—and more recently, political parties. A deepfake-as-a-service business model that he has successfully developed over the past three years.
Jadoun says his company offers a lot more than the simple face-swapping tools available on websites like Deepfakes Web. His products are meticulously produced—with high quality video dubbing, translation and conversational AI.
The Indian Deepfaker is based in Pushkar, Rajasthan, and employs a team of five developers. Jadoun landed his first political project during the 2023 Rajasthan Assembly elections. Ahead of the Lok Sabha elections, he has seven clients from various political parties. (Note: he did not offer specific details of his clients or the work he does for them).
A quick glance at his company’s assignments reveals the range of services such companies offer. TID is creating personalized deepfake videos of a party leader aimed at boosting the morale of party workers. Another involves an AI bot that will call voters in their local language—in the voice of a politician—and hold a two-way conversation about the party and its policies.
The sample we were shown seemed fairly generic: “Hello, this is xxx from ABC party. May I know who I am speaking with? Can I have a few minutes to talk about the upcoming elections?” But an independent consultant Kunal (name changed on request) says the messaging will become more “personalised” closer to the elections: “The names of all voters in a constituency are first put down on an Excel sheet, after which around 30 paise per message is spent on personalising the message, all done internally.”
The true power of AI lies in its ability to transcend linguistic barriers in a country with 22 constitutionally-recognised languages. The BJP has translated Modi’s speeches into Tamil, Malayalam, Kannada, and Telugu—in a big push to up its South India game. Jadoun says AI acts as the great amplifier: "Think of a physical poster announcing the launch of a welfare or policy proposal, and a QR code that when scanned shows the Minister explaining how it works, interactively answering FAQs.” Now, imagine it’s the Prime Minister, and not just any other neta.
The many kinds of deepfakery
Inevitably, Jadoun and his team are approached with any number of unsavoury requests for deepfakes. He’s turned all of them down: “Anything that uses our technology in the wrong direction or in an unfair way that will divide society is considered unethical for us.“ But the requests he describes are chilling in their implications for our electoral process:
“In November 2023, I was approached by a leader to swap his rival’s face onto a sexually explicit video. Despite denying, multiple such requests lie in our inbox,” says Jadoun. ‘Kunal’ offers another—less egregious—example of unethical AI:
There are cases wherein foreign legislators’ faces and voices have been morphed to hail India's legislative prowess… They take the names of Indian members of parliament or candidates as being venerable – indicating the BJP’s success in positioning India as a 'Vishwaguru' internationally.
While we typically associate videos with deepfakes, fabricated call recordings can be equally effective. Voters are targeted with AI-doctored audio of political rivals supposedly arranging for black money for elections or coercing someone to buy votes.
AI technology is powerful precisely because it is flexible—and can be deployed in myriad ways. Jadoun says: “With the advancement of AI, there will be targeted fake news – the kind that allows a candidate to send personalised disinformation to multiple communities, targeting each with their unique insecurities.” In polarised constituencies, even a fake text message on a Whatsapp group can manipulate outcomes.
Doctored videos will have an exponential effect.
AI also allows for an unprecedented level of manipulation. Jadoun was once approached by a Rajasthan candidate who was “caught on video doing something problematic.” He wanted Jadoun to convert the video of his wrongdoing into “a poor-quality deep fake.” The aim was to use the deepfake version to discredit the actual video—“to make viewers believe that the real video was, in fact, a deep fake.” In other words, deepfakes can be used to undermine the authenticity of actual footage—until the voter can no longer distinguish between what is real and what is fake.
A doctored piece of content can be impossible to fight if the timing is exactly right. In November last year, the Telangana Congress shared a 7-second video on X and its official Whatsapp channels of a prominent Bharat Rashtra Samiti leader—KT Rama Rao—saying his “one-point agenda” was to defeat the leader of his own party–K Chandrashekar Rao. As a Congress party worker told Al Jazeera. “[A] normal voter would not be able to distinguish; voting had started [when the video was posted] and there was no time for [the opposition campaign] to control the damage.”
The high price of high tech fakery
The unethical requests come accompanied with eye-watering promises of compensation. As one client told Jadoun: “You do this work and you will never have to worry about money again in your life.” But even the legit kind of AI content is not cheap.
According to The Ken, it costs around Rs 25,000 (US$300) to produce and disseminate a single deepfake video. Estimates provided by a political consultant show that political parties typically assign a minimum budget of Rs 2.5 - 3 million (25-30 lakh) per constituency for digital campaigns—out of a total spend that ranges from Rs 50 - 1000 million (5–10 crore).
Jadoun says he charges more for quality, and his pricing varies depending on the scale and type of request:
A simple voice cloning could cost around Rs 60,000, and around Rs 1 lakh to create a leaders’ digital avatar. At TID, we do personalised video messaging at a starting package of around Rs 3.6 lakh. An additional Rs 12 is charged for videos with different party workers' names.
Political parties with large budgets can push the boundaries of deepfake content—while less well-funded campaigns have to make do with standard voice and video manipulation with limited distribution. Something to keep in mind given the pattern of funding revealed by electoral bonds.
The dark market for deepfakes
Requests for AI content are usually made by PR agencies or consultants affiliated with parties, Divyendra tells us: “It is rare for a party to directly reach out”. That said, scripts for these videos are usually supplied by local party workers—but finally vetted by official reps.
The less ethical requests tend to use private social media platforms. Jadoun says:
Initially, we never understood why some clients would message us and immediately move the conversation onto Telegram, or some private messaging platform. The party or candidates’ PR agency would connect with us through international Whatsapp numbers, surrogate accounts, or anonymous Instagram handles. They would give us a Telegram number to take the conversation forward.
Jadoun has turned down more than 50 such requests from parties and candidates. He has also lost several clients because he insists on labelling his content as ‘AI-generated’.
Not everyone shares his scruples. There are plenty of takers for such lucrative requests. It makes for an excellent side-hustle. “Those using deep fake technology to service unethical requests are normal people, like you and me. They sit in an office, have a day job, and buy online software,” Jadoun says.
Much as other kinds of fake news, these deepfakes will be distributed through surrogate channels. “These surrogates – like Megh News on Twitter – have adopted news personas, allowing the content to seem legitimate, while absolving political parties and candidates of any direct liability”, ‘Kunal’ explains.
As always, ‘WhatsApp University’ will remain among the popular channels of choice. According to Al Jazeera, ordinary citizens “without any public profile” are registered on WhatsApp and used to distribute deep fakes, so as to obscure their association with parties, candidates, consultants, and AI firms. In Rajasthan last year, phone numbers of construction labourers were used to circulate deep fakes.
Deep pockets will make a difference. According to the Indian Express, seven of the top 20 spenders on Meta ran surrogate ads favourable to the BJP between March 17 and 23. There were no other parties on the list.
Where there’s no will…
Both the government and the Election Commission have offloaded the responsibility of monitoring deepfakes onto intermediaries like Meta and X. Given their abysmal track record in battling fake news, social media platforms are hardly up to the task. Companies like YouTube appear to be putting the onus on users—asking them to request the removal of AI-generated fakes.
With an army of deepfake creators at hand, ‘Kunal’ warns the consequences will be writ large on the coming elections: “We haven’t even seen the tip of the iceberg. Deep fakes for this year’s elections will begin coming out around mid-April, close to poll dates.”
