What happens when you build an automated pipeline across five AI tools?
At first glance, this appears to be a single dense illustration. In reality, it is an interactive micro-cinema comprising 100 synchronized, hidden AI video loops. Scaled to its full resolution, the entire artwork spans a 128,000 × 72,000 pixel canvas.
20-Second Teaser
As a kid, I spent hours staring at high-density illustrated books, lifting the flaps of Where’s Spot, finding Wally or exploring again the giant cutaway ship of Richard Scarry. Every room felt like an independent world waiting to be uncovered.
With London Morning, I wanted to test whether modern generative AI tools could recreate that sense of spatial discovery. By designing a modular AI pipeline, I turned a single massive visual into 100 self-contained, interactive micro-narratives.
interactive micro-cinema
Rather than working around model limitations, this project focused on orchestrating AI tools into a structured production engine. Automation handled the heavy lifting, slicing, writing, and animating 100 distinct assets, allowing me to focus on art direction, narrative cohesion, and system design.
Since completing London Morning, I have built four additional interactive canvas projects using this pipeline.
Next step: Adapting this engine for large-scale physical touchscreen installations in gallery and exhibition environments.
Interested in collaborating, hosting an installation, or seeing a demo? Hit me up!
Concurrent Motion Grid (100 Loops)
Behind the scenes
- Base Generation (Nanobanana / Google Flow): Engineered a comprehensive master prompt to generate the high-density, cross-sectional urban base illustration.
- Spatial Segmentation (ComfyUI): Built a custom node workflow to programmatically slice the high-resolution master image into a precise 10×10 grid of 100 individual tiles.
- Contextual Scripting (Gemini): Fed each sliced tile to Gemini via a system prompt instructed to analyze spatial details and write short dialogue loops in the tone of classic London street banter.
- Motion Generation (Veo): Passed each tile alongside its tailored Gemini script into Veo to generate 100 looping motion clips.
- Frontend Orchestration (Google Antigravity): Automated file organization, asset naming, and the underlying interactive JavaScript grid engine for the final web experience.
Base Generation → ComfyUI Slicing → Gemini Contextual Prompting → Veo Loop Generation → Interactive Web Canvas
Behind the Scenes Breakdown
Want to explore the full canvas? This 5-minute recording captures the individual micro-narratives, audio loops, and character interactions across the entire grid.