Gen AI turns New Yorkers' ideas into urban art | Keryc
Imagine you hop on the subway, see a screen, scan a QR and watch your wildest idea for the city come to life in video. Sounds like a movie, doesn't it? But it's real: Google and OUTFRONT launched "Imagine If...", a project that invites New York to co-create urban art with generative models and local artists.
What is "Imagine If..."
It's a civic and artistic initiative that collects visions from residents across the five boroughs and turns them into visual pieces shown on thousands of transit screens. Skyscrapers turned into vertical farms? Neon houses in Brooklyn? You send the idea, an artist interprets it, and models like Veo and Nano Banana help materialize it as video.
Collaboration is key: the public's imagination, the artist's hand, and the compute power of AI.
How it works technically
Spot and send: you see the screen, scan a QR and send a short text with your vision.
Initial generation: the text goes to the generative models (, ) inside the Gemini experience. These models transform descriptions into visual art, typically using multimodal architectures that combine language understanding with image or video generation.
Veo
Nano Banana
Technically, these models usually face two central challenges: temporal coherence (so the video doesn't jump oddly between frames) and stylistic fidelity (keeping the style the artist or community expects). To address these, projects use techniques like:
text-to-image/video models that operate in latent spaces and apply refinement processes (for example, iterative diffusion-like steps) to improve detail and smoothness across frames;
style transfer or artist-guided fine-tuning so outputs preserve a recognizable aesthetic (cut-paper, acrylic, doodles, etc.);
human-in-the-loop pipelines: the artist selects and edits automatic outputs before public display.
In urban deployment there are practical choices: results can be generated in the cloud and pre-rendered for screens for performance and moderation, or produced near real-time if the infrastructure and latency allow. The trade-off is quality and control versus immediacy.
The artists: how human creativity is integrated
Each borough has a resident artist who acts as curator and co-creator. Beyond applying filters or presets, these artists use the generated outputs as raw material: they combine, crop and reinterpret them, then animate the pieces so they reflect a coherent artistic vision.
Practical examples of integration:
Lauren Camara (Bronx) might use the output to create digital collages based on her cut-paper technique.
Subway Doodle (Brooklyn) could insert its iconic characters into generated frames, keeping its creative identity.
This shows a productive working model: AI as ally, not replacement, where human creativity guides aesthetic and message.
Technical and ethical considerations
Who controls the content? What happens to personal data or offensive ideas? In public projects like this, filters and moderation are implemented before display. Transparency about how data is used and the licenses for generated images also matter.
From a technical point of view, there are trade-offs:
Quality vs latency: high-fidelity video requires longer inference time.
Temporal coherence vs creative diversity: making many frames consistent reduces artifacts but can limit unexpected variations that sometimes add value.
Privacy: prompts and interactions must be handled carefully to avoid exposing personal data.
Why does this matter beyond an exhibition?
Because it's a concrete example of how generative AI can integrate into public and cultural spaces: it democratizes creation, opens new ways to approach art, and raises real questions about authorship and governance of creativity.
It's also a practical lab for improving multimodal models in real environments: user traffic, a wide variety of prompts and the need for human curation speed up tool improvements that later apply in other contexts.
Not in New York?
No problem. You can try Veo and Nano Banana from the Gemini app today, experiment and create your own version, even if the physical display is still exclusive to the city for now.
In short: "Imagine If..." turns everyday imagination into public art with a technical flow that mixes generative models, artistic curation and urban deployment. It's an example of how AI can be a tool for community, with technical and ethical challenges worth watching closely.
