000%
Ali Saad / Creative systems practiceSignal acquired
DestinationDestination
Destination / Destination11 / route handoff

All projectsW—11 / 11

HeroCafé

A café hero experiment developed with Codex Astra and video generated in Google Flow, exploring cursor tracking, smooth frame interpolation, and mobile touch.

Try the experiment
Hero Café
OverviewThe buildOutcome

Overview

Role
Creative Developer & Interaction Designer
Client
Personal experiment
Year
2026
View source code

Hero Café explores how a prerecorded character can make a website feel responsive and welcoming. The experience pairs a coffee-holding rabbit with warm café art direction, rounded typography, and a compact button with a liquid-fill hover effect.

The experiment was developed with assistance from Codex using the Astra model. Google Flow was used to generate the character video.

The animation uses every frame from a 10-second source clip. Cursor and touch positions select calibrated poses, while eased motion and optical-flow interpolation smooth the transitions between adjacent frames. The character settles into a recorded neutral pose when interaction ends.

The project is built with vanilla JavaScript, HTML, and CSS. FFmpeg extracts the source frames, and Python with OpenCV prepares the motion data. Small, lossless image strips and a bounded decoded cache make the animation practical to load in the browser without changing the source frames.

This is a creative development experiment in interaction, motion, and browser rendering. Its range of movement follows the poses present in the clip.

The challenge +

The main challenge was making a fixed sequence of poses feel responsive to unpredictable cursor movement. The source needed pose calibration as well as interpolation: smoother timing alone could not correct a target that selected the wrong part of the clip.

Browser delivery introduced another constraint. The complete horizontal sprite was exceptionally wide and a deployed browser reported an image-decoding failure. Splitting the same frames into smaller lossless strips, limiting decoded memory, and pausing the animation clock during loading addressed that failure while preserving the visual output.

Thebuild

01

A character that responds

Calibrated cursor tracking lets the rabbit look toward the visitor. Eased motion and optical-flow interpolation smooth adjacent frames, and leaving the interaction area returns the character to a neutral pose.

All 240 source frames preserved

02

Touch considered separately

The mobile layout uses a dedicated character scene with brief tap responses and horizontal drag tracking. Vertical gestures preserve normal page scrolling.

03

Smaller assets, complete movement

Lossless frame strips load on demand into a four-page decoded cache. Nearby strips are prefetched, and playback pauses at the current pose when it needs another strip instead of skipping ahead.

24 lossless frame strips

Outcome

A responsive café hero that gives a prerecorded character a sense of presence. The rabbit follows the cursor, responds to touch, and gently returns to its resting pose. Smooth transitions connect all 240 source frames, while smaller animation strips load on demand. The live experiment and source code are publicly available.

Source frames
240
Source animation
10 seconds at 24 fps
Frontend
Vanilla JavaScript
Next projectW—01 / 11

Swift Digital Solutions

View project
Swift Digital Solutions
Have something worth building?Start a project
aliihsaad1@gmail.com
InstagramLinkedInGitHubFacebook
© 2026 Ali Saad