When you open CitiScene and see a beautiful 3D diorama of your city under current weather conditions, a lot of technology is working behind the scenes. From real-time weather APIs to state-of-the-art generative AI models, every scene is the product of multiple systems working together in seconds. Here's how it all works.

Step 1: Gathering Real-Time Weather Data

Every CitiScene weather scene starts with accurate, real-time weather data. When you request a scene for your city, CitiScene queries weather data providers to get the current conditions: temperature, humidity, wind speed, precipitation type and intensity, cloud cover, visibility, and UV index.

But raw numbers alone aren't enough. CitiScene also factors in the time of day, which dramatically affects how a scene looks. A sunny afternoon has warm, directional light and sharp shadows. A clear night has cool tones, artificial lighting, and long reflections on wet surfaces. Dawn and dusk create their own distinctive color palettes.

Step 2: Crafting the Scene Description

The weather data is transformed into a detailed scene description that the AI model can understand. This isn't just "rainy day in London" — it's a carefully crafted prompt that specifies the visual characteristics the scene should have.

The description includes information about the weather conditions (light rain vs. heavy downpour, scattered clouds vs. overcast), the location's visual identity (London's Georgian architecture, red buses, the Thames), the lighting conditions (overcast diffuse light, reflections on wet pavement), and the diorama aesthetic (miniature scale, warm studio lighting, tilt-shift depth of field).

Getting this description right is crucial. Too vague, and the AI produces generic results. Too specific, and the scene becomes rigid and unnatural. CitiScene's prompt engineering has been refined through thousands of generations to find the perfect balance.

Step 3: AI Image Generation

With the scene description ready, CitiScene sends it to a generative AI model. These models — built on architectures like diffusion models and transformers — have been trained on vast datasets of images and understand the visual relationships between concepts.

When the model receives a description like "miniature diorama of Tokyo's Shibuya crossing during heavy snowfall at night," it draws on its understanding of Tokyo's architecture, how snow looks under neon lighting, how to create the tilt-shift miniature effect, and dozens of other visual concepts to compose a unique image.

The generation process takes just seconds, producing a high-resolution image that can be displayed on any device. Each generation is unique — even with the same weather conditions and location, no two scenes are exactly alike.

Step 4: Quality and Relevance

Not every AI generation is perfect. CitiScene uses multiple strategies to ensure quality: the AI models are selected for their ability to produce consistent, high-quality architectural scenes; the prompt templates are continuously refined based on user feedback and output quality analysis; and fallback models ensure that if one generation system is unavailable, another can step in seamlessly.

The Three Visual Styles

CitiScene offers three distinct visual styles for weather scenes, each created by adjusting the AI generation parameters:

Cartoon — A playful, illustrated style with bold colors and simplified forms. Perfect for a fun, lighthearted take on the weather that stands out as a phone wallpaper.

Realistic — The signature diorama style that looks like a photograph of a handcrafted miniature model. This is the most popular style, combining photorealistic detail with the charm of miniature art.

Lego — A creative style that reimagines your city's weather scene as if built from Lego bricks. Every building, vehicle, and weather element gets the brick treatment.

Privacy by Design

An important aspect of CitiScene's technology is what it doesn't send to AI models. Only the city name (not your precise coordinates), the weather conditions, and the time of day are used to generate scenes. No personal identifiers, no location history, and no user data are included in the generation process. Your weather scenes are personalized to your city, but your privacy is never compromised.

The Future of the Technology

As AI models continue to advance, CitiScene's capabilities will grow with them. Faster generation times, higher resolution outputs, more detailed and accurate city representations, and new visual styles are all on the roadmap. The goal remains the same: to transform the mundane act of checking the weather into a moment of visual delight.

Want to see this technology in action? Download CitiScene on the App Store or Google Play and generate your first AI weather scene.

The Technology Behind CitiScene's AI Weather Scenes