Generative AI: Spotter Studio

In 2024, YouTube hit two massive milestones: it became the biggest TV streaming platform in the world, beating Netflix with over 12% viewership share, and it was the first streamer to cross 10% of all TV viewing. This explosion in growth was powered by YouTube creators themselves—people like MrBeast, who now has over 400M subscribers and was pulling in an average of 120M views per video in 2024. Behind MrBeast and thousands of other successful channels is Spotter: a company that funds and supports YouTubers in every aspect of their business. In 2024 alone, Spotter invested more than $1B to help creators grow.

But even with all this success, Spotter saw their creators facing a serious problem: burnout. Making videos is exhausting work, especially when you need to keep up with the relentless pace that YouTube audiences demand. Coming up with the right idea or concept—without knowing if it'll actually work—takes a real toll on creators. And then there's optimizing titles and thumbnails, which is both tedious and expensive, but can make or break whether your video lives or dies in YouTube's algorithm.

In this context, Spotter launched Labs, a technology-focused unit dedicated to tackling creators' challenges. I joined Spotter Labs in 2024 as their Principal Designer, with the exclusive goal of spending time collaborating, prototyping, and tinkering on projects directly with YouTubers. Spotter Labs had a significant, unique advantage: unrestricted access to all creators on Spotter’s roster. This access provided unparalleled insights into key areas, such as MrBeast's channel approach to ideation, how successful channels with millions of subscribers integrated long-form content on YouTube, and optimal strategies for titles, thumbnails, and stories across channels of all sizes and audiences. As part of their deal, we spent time with creators every morning, providing a direct channel for experimentation and immediate feedback.

We created over 30 experiments in Labs, all guided by one simple rule: after launching, any experiment that didn’t reach more than 30% Weekly Active Users within a month would be sunset—no exceptions. Some of our biggest successes included:

• Title Exploder: An AI-powered tool to quickly rephrase and test video titles, using exclusive data to pinpoint which versions were most likely to perform well.
• Thumbnail Explorer: A canvas-style interface for rapid thumbnail iteration. For this, we even developed our own image model and collaborated with an artist to craft a unique visual style.
• Story Generator: An AI tool that helps creators build longform stories, allowing them to swap scenes, visualize narrative momentum, and apply classic storytelling principles to drive longer video retention.

The momentum from these successful experiments sparked a larger ambition at Spotter: to build an integrated hub where creators could manage their entire workflow—from the first spark of an idea to crafting the story, packaging the final video, and strategically planning their channel's future.

This vision was about creating cohesion in a fragmented process. The goal was to move beyond individual tools and develop a unified platform that connected every step seamlessly.

That vision became Spotter Studio, the central operating system for YouTube creators—a single, comprehensive home for their creative work.

As one of three Principal Designers on Spotter Studio, I focused on bringing all these separate experiments together into one cohesive experience for creators. After months of working directly with them and seeing how they actually use the tools, a couple of key things became clear:

Creators don't follow one set workflow. Some start with a rough idea that evolves into a full story, then they tackle the thumbnail and title. Others have a specific vision and want to see it as a thumbnail right away. Some juggle hundreds of ideas at once, constantly switching between them. Others prefer to focus on just one or two at a time.

Creators aren't precious about their ideas. One of Spotter's creators put it perfectly: "at this point it's impossible to come up with something truly original. Everyone on YouTube knows it's about putting your own spin on an idea and figuring out what makes your audience click." That's why creators constantly remix each other's concepts and spend so much time hunting for outliers or viral videos they can put their own twist on.

Most creators are more reactive than creative. The creator’s block is real, and going from a blank page to an idea can be incredibly difficult for them. But they're able to tell instantly if an idea will work or not—if a title is going to resonate with their audience, or what tweaks a thumbnail needs to hit big. Their instincts on what is gonna work for them is what makes them successful on YouTube.

Validation is everything to them. One of the big reasons creators love remixing is the built-in safety net—they know the idea has already worked somewhere. At every step, they're looking for any signal that what they're building is going to succeed. Maybe it's performed well in the past. Maybe it's crushing it for other creators. Maybe the data backs up their gut feeling. Whatever it is, they want that confirmation before they commit.

‍

The Studio homepage came directly from these insights. Our first attempt was just a prompt box, which backfired in two ways: it amplified their blank page anxiety, and it didn't show them what Studio actually does—create ideas, stories, thumbnails, and titles.

I rethought the whole thing to give creators multiple starting points instead. For creators who like to remix, we showed them AI-generated ideas based on their channel, trending ideas from other Studio users, and outliers—videos that recently crushed it beyond expectations. For ideas they'd already captured, we displayed predicted performance numbers and a proprietary score showing how likely their video was to outperform both their own average and the industry standard.

And instead of that prompt box, we gave them four clear actions—letting them jumpstart their process however they wanted. They could start with an idea, a story, a concept, or a title, so they weren't locked into one way of working.

‍

We designed everything to be iterative, drawing from what we learned in Spotter Labs. Creators could keep refining their concepts, titles, or thumbnails until they felt right. We surfaced quick actions that our models were specifically trained for, and sprinkled in bits of proprietary data that explained why we thought certain content would crush it.When FLUX dropped in 2024, it completely flipped our approach to thumbnails. Up until then, we were generating detailed sketches that creators would pass along to their thumbnail designers. FLUX let us solve multiple problems at once—we could now create production-quality thumbnails that nailed the creator's likeness and perfectly matched their style. I designed flows specifically for training LoRAs, both for thumbnail styles and for reproducing creator likenesses.

For thumbnails, we hit the limits of what AI models could do with things like text and poses, so we built an in-app editor designed specifically for YouTube creators.

We focused on three main problems. First, the "selfie" faces creators love putting on their thumbnails—we let them either paste one in directly or snap a new selfie through their webcam, which we'd automatically remove the background from and layer onto the thumbnail. The other two were text and shapes—those red circles and arrows that actually drive clicks. We realized pretty quickly that no amount of model training would be faster than just letting creators add these manually, so we built those tools right into the editor. Finally, we also took advantage of FLUX’s inpainting capabilities to provide users with finer ways to tweak their thumbnails.

‍

FLUX also supercharged our Story Generator. Before, we could outline a story structure, but it was still pretty abstract. With FLUX, we could actually visualize scenes and chapters in a way that made sense for these highly visual creators. They could finally see what their story would actually look like on screen, not just read about it.

The generator mimicked how YouTube editors worked on their videos on two levels. At the macro level, creators could see how scenes connected and flowed into each other—basically the entire narrative arc. But they could also zoom way in to the micro level and work through individual scenes second by second, shot by shot. So if they wanted to adjust the pacing of a reveal or rework how a joke lands visually, they could do that without losing sight of the bigger picture. It gave them both the 30,000-foot view and the frame-by-frame control they needed.

‍

During one of our hackathons, I got paired with people from outside Studio—folks from Creator Support and Advertising. We quickly realized something: if we knew what videos our creators had in the pipeline, we could help them land native ad deals and sponsorships way more easily.

We took most of the architecture we'd already built for Studio and spun up a pilot. The concept was simple: let brands peek at the videos our creators were developing so they could bid on them for ads and sponsorships. This accomplished two things at once. First, it made creators more money. But just as importantly, it gave them confidence in their projects—knowing there was already brand interest meant their time and investment would pay off. That confidence became fuel. It gave them the drive to keep creating.

‍

The impact of Spotter Studio on creation was gigantic. In just one year, Spotter Studio helped generate 5 billion incremental views. Videos created through Studio performed 49% better than what creators were posting otherwise. We generated over 10 million thumbnails in the process, cutting creation time by more than 90%.

But the most important part? Creators started creating again—and getting rewarded for it. Videos made through Studio generated $120M in ad revenue for creators that year.

‍