The Prototype That Helped Bring AI Translation to Facebook and Instagram

Thano Sipsis Meta AI translation
Before AI-connected prototyping became a common part of product design, Meta product designer Thano Sipsis helped turn an experimental hackathon idea into a working translation experience that later made its way to Facebook and Instagram.
Written By:
Arundhati Kumar
Published on: 
Updated on: 

In early 2024, product designers were beginning to consider how generative AI could change consumer software. Much of that work, however, was still communicated through static screens, scripted demonstrations, or prototypes that simulated an AI response rather than using the underlying technology.

Thano Sipsis took a different approach inside Meta. He proposed an experience that could translate spoken video content for viewers who did not understand the creator’s language, then explored it as a hackathon project. Instead of illustrating the idea as a sequence of mockups, he wanted people to experience the concept with working AI.

Sipsis built a functional prototype using Origami, Meta’s interaction-design prototyping tool, connected to a Meta AI API service. His contribution included proposing and pitching the initial product idea and doing hands-on interaction design and prototyping. The technical connection mattered: the prototype did not merely show what a translated-video feature might look like. It let collaborators experience AI-generated translation within an interactive consumer-product flow.

“At the time, very few product designers were prototyping directly with these models,” Sipsis said. “The AI coding wave had not happened yet. I wanted to use the actual technology, not fake the response inside a prototype.”

Prototyping the Technology, Not Just the Interface

The prevailing design workflow separated product visualization from the working AI stack. Designers could communicate an intended interface, but a static mockup could not reveal how a model’s output, response time, or occasional mistakes would affect the experience. Those details are especially important in translation, where the technology is mediating communication between people.

Sipsis brought the API into Origami and used it to process video content and return translated text and audio. He connected those capabilities to a high-fidelity Reels experience that worked live: after someone selected the translation control, the prototype translated the video rather than playing a prewritten demonstration.

Prototyping the Technology, Not Just the Interface

“The whole point was that someone could tap a video and hear it in another language right away,” Sipsis said. “It was not a concept deck. The translation actually worked.”

The project explored a simple but consequential ₹product question: could AI allow a creator’s video to reach viewers across language barriers without stripping away the creator’s identity or making the technology the center of the experience?

Sipsis’s specific design contribution was to help make that emerging capability tangible. He proposed and pitched the initial product idea, contributed hands-on work to the interaction prototype, and helped define how a viewer would encounter translated content. The prototype established an early product model for using AI translation within short-form video rather than treating translation as a separate utility.

“I was not trying to show off AI,” Sipsis said. “I wanted people to connect with videos and creators in ways they could not before.”

From Hackathon Prototype to a Meta Product

The concept was subsequently developed into a production feature at Meta. The shipped experience refined the initial design while preserving its central behavior: letting viewers activate translated audio from a Reels interface.

From Hackathon Prototype to a Meta Product

Mark Zuckerberg presented automatic AI dubbing for Instagram Reels during the Meta Connect keynote on September 25, 2024. The demonstration showed translated speech rendered in the creator’s voice, with lip movements adjusted to correspond to the translated audio. Meta subsequently expanded AI translation across Instagram and Facebook.

Extending a Creator’s Reach Across Languages

The shipped capability addresses a large distribution problem for creators. Before automated dubbing, the language spoken in a video limited who could readily understand it. Translation gives creators a way to reach audiences outside their original language without recording separate versions of the same content. Viewers, in turn, can discover people and ideas that previously would have been inaccessible to them.

The feature has since moved far beyond a speculative demonstration. In Meta’s April 2026 earnings call, Chief Financial Officer Susan Li said that more than half a billion users on each of Facebook and Instagram were watching AI-translated videos every week.[1] Meta described the feature as a way to unlock more video inventory, recommend a more diverse range of content, and deliver videos in each viewer’s local language. Its significance is therefore measurable at both a human and platform level: it lets viewers understand creators they could not previously follow, gives creators access to audiences outside their original language, and broadens the content Meta can recommend.

The broader industry has moved in the same direction. YouTube has introduced automatic dubbing for creators, while TikTok now offers AI dubbing, translation, and lip-sync capabilities through its Symphony creative tools. Their adoption shows that the product problem addressed by his early prototype became an important competitive direction for consumer video.

A New Role for Product Designers Working With AI

The translation project also anticipated a change in product-design practice. Today, designers increasingly use code and AI APIs to test product behavior directly. In early 2024, before the current wave of AI-assisted coding tools had made such workflows common, connecting an active model to a design prototype was unusual.

Sipsis continued developing his practice at the intersection of AI and advanced interaction prototyping, using working technology to make emerging product behavior easier to evaluate.

That method changes what a prototype can answer. A traditional prototype can show the intended sequence of screens. A model-connected prototype can expose the actual behavior that a product team must design around. It lets designers and engineers encounter variability, latency, failure states, and user-control questions earlier—before those issues become expensive implementation decisions.

For Sipsis, the live-translation project demonstrated both sides of that contribution. The product concept anticipated a meaningful new use of AI in social video, while the prototype showed a practical way for designers to participate in defining AI products with the real technology in the loop.

“Designers should not have to wait until all the technical decisions are made and then decorate the result,” Sipsis said. “If we can work with the real technology early, we can help decide what the product should actually become.”