The Daring Creatives
Subscribe
Jenny Krakovsky Acted Out Her AI Feature BOSS Before Cloning the Voices
Creator story

Jenny Krakovsky Acted Out Her AI Feature BOSS Before Cloning the Voices

Jenny Krakovsky (@jenny.krakovsky) spent years making commercials and documentaries. Then she made BOSS, a 1960s Liverpool crime film, with AI, reading the main characters' lines herself and running her own performances through voice cloning.

#handle:@jenny.krakovskyNov 18, 2026 · 7 min read · By William Smith

Jenny Krakovsky's route into film started with getting kicked out of school at 14. She describes where she comes from as a "working class background, single parent family, South London," and by 18 she'd made it into film school anyway. That's where she met Colin Young, who founded Britain's National Film and Television School, and he and his partner ended up mentoring her. "They were the people who instilled in me it doesn't matter where you're from," she said in her BBC Radical interview.

She trained in documentary and went on to make commercials and documentaries for Nike, Prada, Adidas, Bentley, VICE, GQ and WIRED, which are all listed on her site. The work paid, but it didn't do much for her. "I was, you know, making some money, making commercials … But I wasn't really fulfilled in what I was doing," she says in the wrap vlog for her film.

So in May 2026 she gave herself 30 days to make a feature with AI, and she vlogged every one of them. It ended up taking 50. The result is BOSS, a crime story set in 1960s Liverpool. To get the voices right, she read the main characters' lines herself, then ran her own performances through voice cloning. The whole diary is public on her YouTube channel, @jennykrakovsky, and she also posts as @jenny.krakovsky on Instagram.

A script that was sitting on her laptop

BOSS started as a script she wrote four or five years ago and never got made, because, as she puts it in her Day 1 vlog, "it's just so hard to get films made in the UK, but also in Hollywood." She told the BBC she had "scripts that are just kind of collecting dust on my laptop."

At the same time, the commercial work that paid for everything was drying up. "The budgets are getting smaller and smaller every year," she told the BBC. Between the two, she decided to be an early adopter, make the film with AI, and use the daily vlogs as her marketing.

The story follows a Liverpool housewife whose husband, Dickie, gets arrested. "She's kind of backed into a corner, and she ends up becoming a crime boss herself," Krakovsky told the BBC. The vlog below opens on one of its scenes, four women in a cafe arguing over whether stealing a footballer's car counts as stealing from a real person.

Jenny Krakovsky — Day 12 of her BOSS diary, "Directing for Subtext," which opens on the cafe scene, on YouTube (@jennykrakovsky)

Reading the parts herself

She spends more time on the voices than anything else, and in her Day 8 vlog, on directing through audio, she explains why she starts from the script. If you wrote it, you already know how every line should sound, because "part and parcel of writing is basically talking to yourself in your head."

That's why she reads each scene out loud herself, every part, in a Scouse accent. "I've been basically doing scene readings myself and playing all the characters and then feeding that into programs that voice clone," she says. The accent is a big part of the reason. She told the BBC, "AI is not able to do a Scouse accent," though she added that it can, just not reliably.

On Day 8 she walks through the whole workflow:

  • She records a scene raw, doing every voice herself.
  • She runs that one recording through the voice tool ElevenLabs once for each character, uncut. Minor characters get ElevenLabs' stock voices, and for the main characters she cloned the voices of Liverpool people she found talking online.
  • She brings the exports into GarageBand and cuts the scene together at the pace she wants, with room to breathe.
  • She chops the finished audio into 15-second pieces, the length of one Seedance 2.0 clip, and feeds each piece in along with the matching lines from the script.

She did try skipping all of that and just handing Seedance the lines, and it didn't work. "It won't be in the right accent, there'll be no uniformity … the performance will be flat," she says. With her own reading underneath, she can generate different camera angles that all carry the same performance.

She directs through the prompt, too. In one scene a solicitor turns down a woman who wants him to take her husband's case, and she wanted him quietly irritated the whole way through. "I told Seedance that he was going to be annoyed throughout the scene and I kept that consistent across the generations," she says, and that one note is where his head shakes and sighs between lines came from.

Jenny Krakovsky — Day 8, "Directing through audio": the finished scene, then her raw read of the same lines, on YouTube (@jennykrakovsky)

The video plays the finished scene first and then her raw recording of the same lines, and she's upfront about where her own skills stop. "I'm not an actor, I have no desire to be an actor," she says on Day 8. What she took from it is that "just by simply having a human input … really changes the output."

When the BBC's Gabriel Gatehouse raised the ethics of cloning voices found online, she didn't wave it away: "There are, and we can go into it for sure. I'm open to talking about that."

She took a lot of criticism for it, and when we asked her about it by email, she said she should have explained it better at the time. Her own reading is the base of every line, so "my voice is the main driver of the performance." The cloning, she says, comes out loose at this point: "the AI approximates and the approximation is not so close." That leaves the voices in the film "essentially inspired by real voices of Liverpool but they're far enough away that they no longer connect in any way with the original voices." She doesn't call it clean, either: "Still not the most ethical thing I've ever done, candidly, but it is what it is."

Shooting it like a real set

Her other big shortcut is borrowed from live-action shoots, where a crew usually films the whole scene wide first and then moves in for the closer shots. In her Day 7 vlog, on coverage, she says "doing exactly that workflow is actually really useful for AI production."

She keeps working on the wide shot until it plays as a scene on its own. Then she hands that video back to Seedance with a prompt she posted in the video's description: "Recreate this scene exactly keeping the dialogue and action exactly the same with no changes. The only change is I want the scene to be broken into medium close ups on the characters who are speaking when they are speaking."

If she runs it four or more times, she says, the close-ups cut together with the wide. It still slips sometimes. In one take she caught a woman in the background who shows up twice.

Jenny Krakovsky — Day 7, "Coverage Workflow": the wide master, then the same scene rebuilt as close-ups, on YouTube (@jennykrakovsky)

The 1960s look

She's a fan of the French New Wave, so that's what she built the look around. In her Day 4 workflow tutorial she explains that the color New Wave films she loves were shot on Eastman Color 5251 film stock, so the stock went into her prompts, along with the Éclair Cameflex, a French camera Godard and Agnès Varda used.

Her photography prompt in that video's description reads: "35mm Eastmancolor 5251 film scan. French new wave 1960s colour grading. 1960s Liverpool in technicolor. Shot on Éclair Cameflex CM3. Documentary aesthetic and composition."

The documentary part matters to her, because when AI footage "looks kind of too glossy, it just doesn't sell itself," as she says on Day 4. She told the BBC she wrote the script in Final Draft and then built her characters and sets as still images in Nano Banana 2. For the sets she worked from archive photos of real 1960s Liverpool terraced houses, the way she would research a documentary.

She didn't end up storyboarding in GPT Image, she told us, and instead "just used Nano Banana 2 to create each start frame individually for each 15 second scene." The video itself came out of Seedance, which she ran on two platforms, Mitte and ACT3.

The finishing was all by hand. "I edited it by hand, I graded by hand and did sound design by hand all in Premiere Pro with no AI," she says, though she adds that "a lot of the look for the grade was baked into the images to start with."

Directing for subtext

Day 12 goes back to one of her first jobs out of school. "One of my first jobs out of film school was working as a camera girl for an actors workshop," she says, and that's where she watched actors work out what their characters wanted and what they were hiding.

It helped with an old problem. Producers had told her that the characters in her earlier scripts all sounded the same. "Basically, they spoke like me," she says.

So for BOSS she gave each of the four women her own outlook. Sally always plays the victim, Jackie is fearful, Pam is the practical one, and Gina, the lead, takes bigger risks as the film goes on. You can see it in her prompts: the Seedance prompt in the Day 12 description tags every line with how it should land, like "(Said with a lot of sarcasm)" and "(Angry and indignant)."

The soundtrack

The songs come from the music generator Suno, mostly Merseybeat-style tracks and 1960s Motown girl-group ballads. In her Day 5 vlog she says that when a track comes out sounding too modern, she adds words like "vinyl hiss, analog, recorded in a basement."

Most of the time she just tells Suno what a song is about. The exception is the scene that introduces Dickie. It's a violent scene and she wanted a love song over it, so she wrote the lyrics herself, a song called "He's Mine Anyway." The full 19-track soundtrack is up on her channel.

Jenny Krakovsky — the soundtrack to BOSS, on YouTube (@jennykrakovsky)

Tools she credits herself

Final Draft · Nano Banana 2 · Seedance 2.0 (on Mitte and ACT3) · ElevenLabs · GarageBand · Suno · Premiere Pro (edit, grade and sound, no AI)

From her Day 4, 5, 7 and 8 vlogs, her BBC Radical interview, and her answers to us.

After the wrap

She finished the color grade the same night she recorded the wrap vlog, which went up on August 31. She says the credits have almost nobody in them, since she did everything herself, while the special thanks run long with people she met online during the challenge. The vlogs were the whole marketing plan: "I haven't spent a penny on a PR campaign. I am the PR campaign."

The plan had been to put BOSS on YouTube, until an offer came in that could get it into theaters. Roger Avary, co-writer of Pulp Fiction, has been championing the film, and his company, General Cinema Dynamics, is distributing it. "We have a theatre chain on board and it'll be in selected theatres in the USA in December 2026," she told us, and she calls that the first rollout.

Meanwhile she's lining up the next one. In the wrap vlog she invites viewers to co-direct her next feature by posting their best AI film in the comments, and she says that film "is going to break a world record." We asked what the record is. "You'll have to wait and see!"