The animation industry is at a critical crossroads. Amid shifting priorities, job anxiety and the explosion of generative AI since 2023, many creatives feel like the ground is slipping beneath their feet.
Online backlashes against AI-assisted projects are frequent and fierce, as the Amazon GenAI Creators’ Fund controversies showed. Many artists are determined to avoid the technology entirely, but one animator has taken a different approach, working directly with one of the world’s biggest AI research labs.
Having cut her teeth as a CG generalist on Netflix projects before landing a role as a story artist on Pixar’s Inside Out 2, Connie He made an unexpected leap to Google DeepMind. There, she’s directed Dear Upstairs Neighbor, an animated short built using original art as references for a unique generative AI animation pipeline (see the trailer below).
I caught up with Connie to learn why she decided to cross the divide, how her team built a pipeline that respects human craft, and why she believes that, rather than producing generic AI slop, Generative AI can actually help make animation more diverse.
Connie admits to experiencing a period of profound internal conflict when tools like Midjourney and DALL-E first emerged, but she seeing the immediate impact on the sector, she decided that ignoring the technology wasn’t an option.
“The motivation to start jumping in is because I feel like if this technology is going to keep developing, artists need to stay in the conversation rather than shy away,” she told me. “We need to build a bridge and have this conversation with the people who are building the technology, and build a mutual understanding on what creativity is and what artists need.”
For her, the breakthrough moment came from an unlikely source: a vintage interview in which the Icelandic musician Björk discussed people’s views of electronic music. “People always talk to me like computer music doesn’t have a soul, but it doesn’t have a soul because you didn’t put soul into it.” the animator recalls Björk saying. “That moment was a kind of enlightenment moment for me. The question becomes: how can you inject yourself into it? It’s a tooling question.”
For those who worry that AI means typing a text prompt and letting a machine do the work, Dear Upstairs Neighbor demonstrates a different blueprint.
The short tells the story of Ada and her trouble sleeping due to noisy neighbours. It’s a simple tale with some surprising twists told through a visceral art style that seeks to capture the character’s increasing desperation.
The project rejected text-to-video generation entirely, focusing instead on image-to-video pipeline using hand-drawn art and purpose-trained machine learning models.
Every character design, storyboard and piece of concept art was created by human artists. Connie herself painted a series of abstract acrylics on paper that were scanned and used to train custom models, or LoRAs. These were then used in a hybrid 2D/3D animation process along with Maya, one of the best animation programs. The result is a complex, textured “living painting” look that traditional CG animation pipelines simply wouldn’t be able to replicate.
“In computer graphics, it’s always hard because everything will look CG – it’s how the rendering systems are built, and things move in a CG way,” Connie notes. “But the machine learning model is able to pick up the design rules we give it, something you cannot get with computer graphics. Could you do it with traditional 2D painting? Technically yes, but it would probably take you 10 years.”
Traditional CG animation pipelines are heavily siloed, with artists handing work off to render farms. Connie says that at DeepMind she’s working hand-in-hand with software engineers to build tools that prioritise artistic control over automation.
When the team ran into a classic AI hurdle flickering output, she collaborated with a researcher to build a custom system using masks exported from Maya to stabilise the character’s face, while allowing the hair and backgrounds to retain a gritty, painterly flicker when desired to reflect Ada’s emotional state.
“A lot of the time, it was the researchers who inspired me,” she says. “I couldn’t even imagine this could be possible, but they understood my intention and built a system to make it possible. Now I can see this type of effect is possible, how can we use that to maximise our storytelling?”
Image 1 of 5
There is an elephant in the room, of course. Connie is aware of the ethical anxieties around the use of AI. She used models trained on her own art but recognises the copyright concerns around the foundational models that underlie them
“I’m hoping there will be some legal regulation coming in because that area is blank right now,” she says. “Without legal framework, it’s going to be hard to pursue this conversation in a thoughtful and respectful way.”
As for whether major studios like Pixar could embrace AI animation, she remains realistic. “It will take time for both the technology tool side and the people running the studios to figure out what’s coming. It will be chaotic for a while, but eventually, there will be good people working to make it less chaotic.”
Ultimately, she views AI not as a replacement for the artist, but as an unfinished medium that needs artistic guidance, and which could, if used astutely, unlock new creative possibilities for animators.
“To me, it’s like a new, fresh box of paint. Watercolour is a medium, oil painting is a medium, computer graphics is a medium. Right now, this new box of paint only has a few colours in it for artists to use. There are a lot of blanks. But when that expands, I’m hoping artists can use them to create stuff that used to not be possible.”
