Local Canvas is a collaborative project within Triple Innovation. The app is designed to transform videos into living artworks. Users upload a video, and our back-end AI identifies objects in the footage—such as people or cars. These objects can then be styled with specific artistic techniques, like Pop Art or a Gustav Klimt-inspired look, while the rest of the scene remains untouched. This creates a unique hybrid visual experience where AI-styled elements coexist with real-life surroundings.
Montage of Local Canvas Outputs
A demonstration of how Local Canvas detects objects in videos, applies artistic styles to them, and integrates them seamlessly into the scene.
Deep Dive into Local Canvas
A detailed walkthrough of how Local Canvas works, from object detection and image segmentation to applying AI-generated artistic styles and reconstructing the final video output.
In addition to applying predefined artistic styles, we also experimented with enabling users to create custom models based on their own drawings or photos. For instance, a user could draw a car, upload it, and the app would generate a model that applies their custom style to real cars in videos. This shifts the AI’s role from a passive tool to an active artistic collaborator, allowing users to remix reality in entirely new ways.
Custom Model: César Manrique-Inspired Style
An experiment where Local Canvas uses a custom LoRA trained on the artistic style of César Manrique, applying the style to a car detected in the video.
Custom Model: Hand-Drawn Car Style
A simpler experiment where Local Canvas uses a LoRA trained on a single hand-drawn image of a car. This custom style is then applied to a real-life car in the video.