Inside DeepMind’s vision for creative control - Think with Google - Google Business Profile

In the rapidly evolving landscape of generative AI, the conversation is shifting from what models can do to how humans can guide them. Douglas Eck, senior research director at Google DeepMind, is at the heart of this transition. In this exclusive interview, he explores the leap from simple prompting to sophisticated creative directing, the importance of preserving the craft of artistry, and how professional-grade tools are redefining industries from filmmaking to marketing.

Think with Google: Out of all the recent launches, what has you most excited right now?

Douglas Eck: I’m excited about Gemini Omni, no question. It’s a model that will pull together and fuse all the modalities, so you can create anything from any input, starting with video but soon including audio, images, text, and more. But the real value proposition isn’t just technical. It’s about putting control back into the hands of artists so they can get exactly what they need from our models.

In the early days of AI image generation, you couldn’t control the models much at all. You’d set some random numbers and just hope you’d get a face or something recognizable. Then text prompting happened, which allowed us to use language. Now, we are moving into a stage of true controllability. For a creator, control is more important than pixel quality, resolution, or lighting. It’s the difference between pressing a button and hoping for the best versus actually directing the output. It puts control back into the hands of artists to get what they need from our models.

Think with Google: By simplifying how we guide AI outputs, are we democratizing creativity?

Eck: There’s a difference between democratizing creativity, and democratizing access to tools. I want to democratize access to tools but creativity itself is hard. Creativity should be hard. Creativity is about working with your mind and ideas to hone something you want to share with others. It is intrinsically not easy, and we aren’t trying to make it easier. Making creativity easy is exactly what leads to AI slop. We want to preserve the craft, enabling creatives to direct properly rather than just generating filler.

Think with Google: You’ve noted that the film industry has been faster to adopt AI than other creative sectors. What can marketers learn from that?

Eck: The film industry already has a proxy for AI: computer graphics (CG). Filmmakers are used to working with special effects (SFX) shops where the craft and artistry are preserved. The filmmaker knows how they want the shot to look, and they give very specific directions and ship the existing assets off to the SFX artists and wait a month or so to get the rendered shot back. But now, AI models can remove that lag time and tighten the creative loop.

I remember talking to a famous actor and producer about how these models work. When I told him he could get results in minutes or even seconds rather than waiting a month for a CG shot, he just lit up. It enhances the creative process. With AI, the filmmaker can be in the room with their director of photography, with the actors, with the special effects artists and experiment and create and realize their vision together.

Think with Google: Can you share an example of AI achieving something that was previously impossible?

Eck: Our crew worked on a short film called Dear Upstairs Neighbors, directed by Connie He, who came to us from Pixar. She physically scanned her own paintings to fine-tune a generative model.

The surprising part was that the model not only learned the colors and the textures of her style but the underlying geometry. Most video models are trained on media from physical cameras, which have slight lens distortion. Our concept art was in a painterly, 2D perspective with perfectly straight X and Y axes. The model learned to generalize that style and apply it in completely new prompts, like creating a barbell that remained perfectly straight, something very difficult to achieve in traditional CGI.

Another surprise was the main character’s ponytail. If you have a ponytail and spin around, at certain points your head blocks the ponytail from the camera’s view. But for this film, it was an artistic choice to have the ponytail always visible as a persistent visual cue. With a traditional 3D animation model, keeping that ponytail visible would have required elaborate and laborious rigging and likely redrawing individual frames. But our model found a way to “pop” the ponytail around so it was always visible, regardless of the rotation. It created a look that would be impossible to rig in 3D without immense effort.

Think with Google: There’s a parallel between how an artist needs AI to maintain consistent style and how a marketer needs AI to follow brand guidelines. How do improving AI systems make this possible?

Eck: We’ve used machine learning in our recommendation systems for years. On YouTube, they relied on collaborative filtering: If Jane liked this and you’re like Jane, you’ll like this. It’s worked incredibly well. But with advances in AI, we can do much, much better.

I use the analogy of a sommelier. No sommelier ever says, “Please share with me your last 500 wine purchases.” They don’t need that, because there’s intelligence at play here. They can understand your intent and context in the moment. Perhaps AI-guided advertising could be more like that, a more helpful conversation that lets a brand shape its message for different audiences, like moving a campaign from one market to another, while keeping the core brief intact and keeping all the messaging and assets on brand.

Think with Google: What does the future of agentic AI look like when it comes to the evolution of these creative tools?

Eck: Agentic AI is the next frontier that everyone is working toward. In my mind, the major research challenge is getting different agents to work cooperatively. I compare the current state of multiagent systems to “swarmball” — like 7-year-olds playing soccer where everyone just runs to kick the ball. It’s chaotic and often counterproductive.

In the future agents might have specific roles, and all contribute to a living brief. You might have one agent that represents the client, another that ensures brand consistency and safety, and a localization agent that understands which formats and channels work best in specific regions. If they work together, you get a living campaign that is constantly updated and refined.

Think with Google: What’s the next frontier for this technology in 2027 and beyond?

Eck: The physical world! We need world models that are accurate down to the millimeter or the micron. That’s what’s needed to make robots that can build homes and for manufacturing. And that’s already happening with Waymo right now, which are practically automotive robots moving through the world.

In the arts, I’m excited about experiential technology. We’ve already done projects with Sphere in Las Vegas, including The Wizard of Oz at Sphere. My dream is to see a musician or actor on stage using the entire room as an extension of themselves, controlling the environment with their emotions and voice in real time. It’s not about hyperpersonalization. It’s about contextualization and creating an insane, responsive audiovisual experience.

It is such an exciting time. And for marketers, my message is really blunt: The opportunity cost of sitting on the fence is way higher than just jumping.