MOMNT

A guest at a live event does not want to become a prompt engineer. They want to participate quickly, recognize the source in the result, and share something that feels specific to the room.

The design challenge was to hide the machinery without making the experience mysterious: set expectations before the capture, make waiting legible, reveal useful results progressively, and let the guest retain the editorial decision.

Behind the scenes, OpenAI, Gemini, and Runway receive provider-specific versions of the same creative direction and return independently. Guests simply see Portrait 1, 2, and 3, so they choose according to the result—not the provider behind it.

These three submitted moments span three events and three very different Looks. Each pair shows the untouched camera capture beside the image the guest chose for the live gallery.

The challenge was not simply generating media; it was turning probabilistic models into an experience guests could trust without instruction.

Each provider job has its own progress, success, moderation, and recovery state. Results reveal as they arrive instead of waiting for the slowest model. A partial failure does not erase successful work. Guests see a calm editorial choice; operators can inspect the provider, model, prompt version, latency, cost, and structured error behind every attempt.

That separation lets the public experience stay light while the production system remains observable and adjustable.

MOMNT spans several people, screens, and moments in time. Hosts configure the event name, welcome, allowed Looks, schedule, logo, message style, gallery, QR, and presentation behavior. Guests use their own phones. The room-scale presentation polls for new submissions and group images, synchronizes against server time, pre-buffers scheduled video, and lets late joiners recover the current state.

Designing that system meant treating coordination—not any single screen—as the interface. A host can set the tone, a guest can contribute without training, and the room sees the collective result arrive together.

MOMNT first grew through private hotel and venue events. Those live deployments exposed the details a concept prototype misses: throughput, moderation, recoverability, guest comprehension, operator controls, export, and the choreography between personal phones and a shared display.

We used that learning to open the platform to anyone. A host can now sign in, create and manage an event, choose Looks, configure the experience, share its QR code, and manage the results without our team operating the activation. MOMNT evolved from a bespoke creative service into a reusable creator tool.

By September 2026, the production admin contained 41 event records, 547 capture-session records, and 11 completed generated videos. These are total system records rather than counts of unique clients or users. In a four-week June–July operating window, MOMNT processed 824 image render records across 271 capture sessions, producing 764 successful outputs across the three provider paths.

These production traces became design inputs. They exposed latency, cost, partial failure, model drift, and operational edge cases—and showed where the experience needed to become more resilient outside a controlled demo.

MOMNT reinforced that generative video is an interaction medium, not only an output format. The designer has to shape the full temporal experience: how intent enters the system, how waiting feels, how uncertainty is communicated, how identity persists through transformation, and how the final motion result becomes legible and shareable.

The strongest product decision was to keep the human moment at the center. Capture first. Let the system do the complex work. Return meaningful choices. Then use those individual choices to make something the room owns together.

Create a MOMNT