Kelly Boesch spent 17 years at IMAX doing graphic design, film production, and marketing. She built development decks that convinced studios to fund projects. She understood production rhythm, visual grammar, large-format cinematic work — in a professional context, not from watching YouTube.
Three years ago she picked up AI image generation. She didn't dabble.
She has 2-3 million followers and 60 million views a month as of mid-2026, a debut album on Nettwerk Music Group, and a 2026 gallery exhibition in London. That trajectory is interesting. But the process behind it is what I keep wanting to understand.
Start with an image, not a prompt
Most people approaching AI video do it text-first — describe the scene, generate, hope. Boesch works the opposite way. She generates a still with Midjourney, gets it compositionally right, then animates from that image. The control she keeps over aesthetic and framing is deliberate. Text-to-video is closer to a lottery; image-to-video is closer to art direction.
For animation she pulls from Runway ML, Pika, Higgsfield, and VEO3, depending on what a scene needs. Each tool has different behavior. She moves between them.
Her framing of the edit layer, from an interview about her process, is the clearest articulation of what she's actually doing: "You can't just let the AI do the work — the editing, color correction, and sound mixing are where the human touch turns a clip into a story." That's a methodology, not a disclaimer.
Adding Seedance, July 2026
In late July she added another model to the rotation, and reached it through a tool already in it: "using #Seedance 2.0 in #RunwayML," as she tagged it. On July 25 and 26 she posted two tests within a few hours of each other, running the model in each direction.
The first went text-to-video. She'd been watching the trailer for The Odyssey and wanted to see whether Seedance could reproduce its shot of a massive whirlpool at sea. Her own account of the result: "It came out insane. This app is just incredible. These were pretty basic prompts."
The second went image-to-video — a Blade Runner-style piece she described as "just a bunch of animated images made for fun." She labeled both as departures rather than a change of method: "Very different than what I normally do. Every now and then I like to experiment a little with different things to see what I can get."
Three days later the model was in a finished piece. "The Art Of Being Strange," posted July 29, is the story-driven video she'd been wanting to make, and she credits the new tool plainly: "I have been practicing with Seedance a bit and used it for this one." Her note on what proved difficult is the part worth keeping — "The hard part is trying to direct the things that happen in the scene." Which is the same problem the image-first method exists to solve, showing up again one tool later.
Then add music
Her debut album — Fairytale, April 2026, Nettwerk, 14 tracks, 49 minutes — came out the same week as her London show at W1 Curates. She wrote the lyrics. She used Suno to produce the music under her direction. The label and the imprint (Gelsomina Studios) are real.
This is the part that changes the conversation about what she's making. Directing a generative music tool through your own lyrics to produce a cohesive album that lands on a real label is not a random output. That's a music production pipeline she owns.
You can argue the tools are doing the heavy lifting and she's curating the results. That's a fair read, and I've had the same thought. But curation at that level of intentionality is close to authorship. Most film directors don't operate the camera either.
What the IMAX years actually did
The 17 years matter here. When Boesch talks about color correction, edit pacing, and sound mixing as the human contribution, she's not reaching for concepts she recently learned. She built those instincts professionally. She arrived at these tools already knowing what a cinematic image is supposed to feel like.
That's not a small thing. A lot of AI art fails at the sensibility level — technically generated, but visually incoherent. Her training gives her a baseline the tools can't provide.
The April moment
The timeline is worth noting: Fairytale drops April 10. W1 Curates London opens April 10. She speaks at TED2026 in Vancouver on April 15. The talk was called "Art, Music, AI and The Joy of Creativity."
I don't know whether all of that landing in one week was strategic or just how it fell. Either way, it marks a shift — from "AI creator with a big following" to "artist with gallery and conference stage recognition." Those are different categories, and she crossed into the second one.
Her stated position, and what she's making now
Boesch's line on the tools has stayed the same: AI augments creativity, it doesn't replace it. The workflow is where she puts that into practice — the still she composes before anything moves, the lyrics she writes herself, the editing and color and sound she keeps by hand.
The music side keeps moving with it. Several of the summer's videos are built on her own older songs, reopened and rewritten rather than replaced: "Marshall Altman and I reworked the lyrics a bit and updated the song and it works really well now," she wrote under the July 29 piece. "I like to use some of my old songs again from time to time."