Reface—Climbing to the Top of GenAI (Expert Talks)

Ever heard of an "AI generalist"? Probably not—yet. But mark my words, it's a role that's set to explode as tech companies start fully exploring the possibilities of AI and ML.

As you know, we chat with friends and colleagues from fellow product companies to share our mobile app experiences and pick up some new tricks. This time, we caught up with Ihor Levenets, a Product Manager at Reface, a trailblazing content creation company from Ukraine. One of their main products, the Reface mobile app, has been downloaded over 250 million times within just two years of its launch.

A Look Inside Reface and Their Products

Reface is a company that's taken the lead in GenAI, with eight standout products to its name. Since 2018, they’ve been creating AI tools that push the boundaries of content creation, making it more inventive and within everyone’s reach.

Their Reface app has been a huge success, with over 250 million downloads in just two years. Reface is committed to bringing advanced machine learning to everyone, consistently launching new AI-powered products every year. Today, the company's portfolio includes:

Reface

A mobile app that allows users to replace faces in videos and photos, voice-over content, and generate AI portraits from multiple selfies. Why has this app become such a hit? Well, who wouldn't want to see themselves as a movie star or a famous meme character? It's like a digital costume party where everyone can join in the fun without leaving their couch.

Revive

A mobile app for animating images. It allows users to instantly edit and animate photos, memes, or art portraits. Users love animating images because it brings their pictures to life in the most delightful ways. Imagine taking an old family photo and making everyone in it smile and wink at you—it's a fantastic way to connect with relatives you haven't had the chance to meet or see in a while.

Restyle

A mobile app for creating AI images and changing the style of images and videos. This app is a fantastic tool for content creators because it offers endless possibilities for artistic expression. Whether you're a professional designer or just someone who loves playing with visuals, Restyle can transform your ordinary pictures into extraordinary pieces of art.

Letsy

A mobile app for trying out different outfits. Letsy is a great choice for stylish people of all ages and budgets because it lets you experiment with fashion without spending a dime. From casual looks to high-end ensembles, you can see how different outfits suit you before making a purchase. It's like having a personal stylist right in your pocket.

InkAI

A mobile app for generating and trying out tattoos. InkAI is fantastic because, let's face it, getting a tattoo is a big decision. With this app, you can try out different designs and placements before making the commitment. As the saying goes, "You only live once, but with InkAI, you can test-drive your ink."

Memomet

A mobile app for creating Ukrainian memes. Memomet is a great project because it focuses on improving mood and easing stress. In today's fast-paced world, a good laugh can go a long way, and creating memes is a fun way to share humor and cultural references that resonate with the Ukrainian community.

Metahead

A technology for the neural rendering of 3D images based on 2D photos. Metahead is a crazy idea in the best possible sense. Turning simple 2D photos into fully realized 3D models is like something out of a sci-fi movie. It's a testament to the incredible advancements in AI technology that make the impossible possible.

Unboring

A web resource for generating content that unites all the above-mentioned apps. Unboring lives up to its name by offering a platform where creativity knows no bounds. Whether you're using Reface, Revive, Restyle, Letsy, InkAI, Memomet, or Metahead, Unboring makes it easy to create and share exciting content across different mediums.

GenAI–What It Is and How It Works

Based on the amazing work Reface is doing, let's take a closer look at what GenAI is and why this technology has become so widely used in the entertainment industry. GenAI, short for Generative AI, refers to AI systems that can generate new content, whether it's images, videos, text, or music. These systems learn patterns from vast amounts of data and then create unique outputs based on what they've learned.

When developing a GenAI product, the process is pretty similar to any other product development. You have to think about the user's needs, ensure the product is useful, and make sure it has a viable business model. The main difference is that you're working with artificial intelligence, which adds a layer of complexity. Teams have to interact in specific ways and timelines for new features or technologies might be longer compared to regular products. But from a product development standpoint, it's quite similar to any other tech product.

In the entertainment industry, GenAI is a game-changer. It allows creators to produce content faster and with more creativity than ever before. From generating realistic avatars to creating stunning visual effects, GenAI opens up a world of possibilities. Companies like Reface are pushing the boundaries and making this technology accessible to creators everywhere.

But what exactly is an AI generalist, you may ask? Let's break it down.

Meet the AI Generalist

So, who is an AI generalist? Great question! It's a pretty unique role that's just starting to gain traction in the tech world. Think of an AI generalist as a jack-of-all-trades in the machine learning (ML) space. They can handle the whole ML pipeline, even when the tools are still a bit rough around the edges. These folks can work with engineers to fine-tune models and get everything ready for launch. It's all about taking the nitty-gritty of ML and turning it into something usable and awesome.

At Reface, the AI generalist role has been a natural evolution. Most of their AI generalists started out in different roles and grew into the position internally. They didn't post specific job openings for AI generalists or production engineers; instead, team members adapted and evolved into these roles. Many started in the content team, picking and curating content for the mobile app.

What's cool about the AI generalist role is that there's no set background needed. Whether you're a product manager, an engineer, a designer, or something else entirely, you can grow into this role. It's all about having the right mindset and a willingness to learn. At Reface, people from various backgrounds have become AI generalists, showing just how versatile and adaptable you need to be in this field.

The AI generalist is the ultimate multitasker in the ML world. They're the ones who can take rough ideas and turn them into polished applications. Flexibility, creativity, and a readiness to tackle new challenges are key. As more companies start to see the potential of AI, the demand for AI generalists is only going to grow. So, if you're looking for an exciting, evolving role in tech, keep an eye on this one.

Finding the Right Fit in Product-Market Fit

Product-market fit is a term that’s getting thrown around a lot these days. Simply put, it means that your product meets the needs of a specific market segment and users find it valuable enough to keep using it. For mobile apps, this often translates into how frequently and consistently users engage with the app—also known as retention.

Different mobile apps and niches have varying levels of usage frequency and user engagement. Let’s take Reface's InkAI app as an example. InkAI is all about generating tattoo designs. Now, getting a tattoo isn’t something people do every day, so naturally, the retention rate might be lower for casual users. But InkAI caters to two segments: those looking to get a tattoo and tattoo designers or artists. The latter group might use the app frequently to generate designs for their clients, possibly even daily.

Here’s where product-market fit comes into play: understanding the different usage patterns of these segments. For casual users who are thinking about getting a tattoo, they might interact with the app infrequently. They could play around with it, generate a design, and then not use it again for a while. This lower frequency of use doesn’t mean the app isn’t a good fit for them. It just means their usage pattern is naturally less frequent.

On the other hand, tattoo artists and designers may have a higher retention rate because they need to generate designs regularly for their clients. This means they are likely to engage with InkAI more consistently.

Long-term retention is key for Reface. Even if casual users don’t engage daily, staying subscribed and occasionally checking back can indicate a successful product-market fit. It's crucial to recognize that not every app needs high daily usage to be considered a success. For apps like InkAI, focusing on how often users return over the long haul provides a better measure of success. Long-term engagement, where users keep the app installed and return periodically, shows that the app continues to provide value, which is a true indicator of product-market fit.

Reface looks at both short-term and long-term retention to better understand how well InkAI meets the needs of its different user segments. This approach ensures the app continues to be valuable to its users, helping to tweak the product to better fit the market, and ultimately driving sustained engagement and success.

Why and When Adding More Use Cases Work

Product-market fit is crucial for mobile apps, and one effective strategy to improve retention is by adding new use cases. This approach is especially useful when users don't engage with the app frequently.

The concept of "natural frequency" helps here. According to Reforge, natural frequency refers to how often users naturally return to your app. If users engage frequently, like daily or weekly, your app is in the "habit zone." But if the natural frequency is lower, like monthly or less, adding new use cases can boost engagement.

Take Reface’s InkAI app, for example. InkAI generates tattoo designs, but people don't get tattoos every day. Tattoo artists, however, might use it more regularly. To increase engagement, Reface could add features that encourage users to return more often.

Zillow provides a good example. Zillow is a real estate app where people search for homes to buy or rent. Since buying a home is an infrequent activity, users don't naturally engage with the app often. To increase engagement, Zillow introduced a feature called Zestimate, which provides monthly updates on the estimated value of users' homes. This gives users a reason to check in regularly, even if they only buy or sell homes infrequently. Similarly, InkAI could add a feed for browsing and sharing tattoo designs, encouraging users to visit the app more often for inspiration.

New use cases can also make your app more versatile. Reface initially focused on a meme editor. By adding features like a feed for browsing and sharing memes, they boosted their 30-day retention rate. Users are more likely to visit the app regularly to scroll through new content.

It's important to ensure that new use cases are relevant to your existing audience. If the added features align well with what users are already interested in, they are more likely to engage with them. For example, InkAI users might appreciate a feature that helps them find local tattoo artists or shows trending tattoo styles.

However, adding too many diverse use cases can dilute your app's core value. Users might get confused if the app tries to do too much. The key is to enhance the core experience without straying too far from what made the app successful.

A/B Testing–Common Pitfalls to Avoid

A/B testing can be a powerful tool for optimizing GenAI apps, but it's not always as straightforward as it seems. Here’s a look at some common pitfalls and what to avoid when running A/B tests specifically for Generative AI applications.

As we know, A/B testing, or split testing, involves comparing two versions of a feature or content to see which one performs better. For GenAI apps, this could mean testing different AI-generated outputs, features, or user interactions to see which version drives more engagement or satisfaction. The common pitfalls to avoid in this case include:

1. Running A/B tests for everything. Not every change needs an A/B test. Sometimes, you can use past data and analytics to make decisions. For instance, checking historical data to see if there’s already a pattern or trend can save you time and effort.
2. Ignoring group homogeneity. Make sure your groups A and B are homogeneous. This means that the characteristics of users in both groups should be similar. If group A has 30% men, group B should also have around 30% men. This applies to other variables like age, location, and usage patterns.
3. Overcomplicating tests. Sometimes simpler tests can give you clear answers. For example, if you're testing a new AI-generated image feature, a straightforward comparison between the old and new versions can reveal useful insights. However, it’s crucial to understand what you are testing and why.
4. Not accounting for multiple exposures. Ensure that users are not exposed to both versions of your test. If a user sees both versions, it can skew your results and make it difficult to determine which version is truly better.
5. Relying solely on traditional statistical methods. Traditional methods like the t-test might give false positives, especially with large sample sizes and small conversion rates. Consider using Bayesian statistics, which can provide more reliable results for A/B tests in GenAI apps.

Consider Reface's experience with their push notification tests. They were testing different chains of mobile push notifications to see which one was more effective. However, they encountered a significant issue: users were being exposed to both A and B versions of the test. This exposure led to confusing results because it wasn't clear which version was driving user behavior. The test results were unreliable until they fixed the issue by ensuring that users only received one version of the notifications.

‍Be clear about what you're testing. Have a specific hypothesis for each test. For instance, if you’re testing a new feature that generates AI-driven content, understand whether you’re looking to improve user engagement by changing the style of the content or by improving the AI’s accuracy.

‍Use historical data. Before running new A/B tests, analyze past data. This can help identify if a test is necessary or if a simple change based on previous trends would suffice.

‍Model tests using historical data. Consider using Monte Carlo simulations to model your A/B tests on historical data. This means running numerous simulated tests based on your existing data to understand potential outcomes and biases before implementing the actual A/B test. It helps in identifying and mitigating risks and uncertainties.

What's New and Exciting in Gen AI

The world of GenAI is evolving at a breakneck pace. It feels like every day there's something new and groundbreaking happening. Keeping up can be a challenge, but it’s also incredibly exciting to see the possibilities unfold.

From Text to Visuals and Beyond

While many people are familiar with text-based AI models like GPT, the realm of GenAI is expanding far beyond that. At Reface, there's a big focus on visual content, and the advancements in this area are truly impressive.

Video Generation

One of the most fascinating developments is in video generation. Creating videos based on prompts is becoming more sophisticated. Companies like Runway ML are pushing the envelope with this technology, making it possible to generate high-quality videos automatically. This has huge implications for creatives, enabling them to produce engaging content much faster and with less effort.

Audio and Music Generation

The field of audio and music generation is also taking off. For instance, platforms like Suno allow users to write lyrics and then generate a song in the chosen musical style. This technology isn't just a novelty; it's being used to create real, engaging music. 

Recently, there was a trend on TikTok where people were manually creating songs in the style of bands like Linkin Park. Now, AI can do this automatically, making it accessible to everyone.

AI Dance Animations

Another fun and impressive technology is AI-driven dance animations. Imagine taking a static photo of yourself and turning it into a dancing video. That’s exactly what AI Dance can do. This tech takes your photo and animates it to mimic a dancing video, making it look like you’re the one dancing. It’s incredibly viral and visually stunning. We’ve seen teams use it to create entertaining and engaging presentations, showing off their monthly progress in a fun way.

Multi-Modal Integration

The future of GenAI looks even more promising with the integration of different media formats, known as multi-modal AI. This means combining text, audio, and visual elements to create rich, interactive content. 

For example, you can input your interests and hobbies, and the AI will generate a personalized song that fits your style. This blending of media types creates a more immersive and customized experience, showcasing the true potential of GenAI.

These innovations are just the tip of the iceberg. As AI continues to advance, every type of media will become more intertwined with AI technology. Whether it's video, audio, or text, GenAI is set to revolutionize how we create and consume content. 

The possibilities are endless, and the future looks incredibly bright for GenAI.

Open-Source Models and the New Wave in AI

Open-source models are making waves in the Generative AI (GenAI) community, and they’re catching up fast to their proprietary counterparts. If you’ve been around long enough to remember when GPT-4 was released just over a year ago, you might be surprised to learn that today's top open-source models are already outperforming GPT-3.5.

What are open-source models?

Open-source models are GenAI models whose code and architecture are publicly available for anyone to use, modify, and distribute. This openness fosters innovation and collaboration, allowing a wide range of developers and researchers to contribute to and improve the technology.

The rise of open-source in GenAI

The rapid development of open-source GenAI models is nothing short of impressive. While companies like OpenAI and Anthropic are known for their advanced proprietary models, the open-source community is quickly catching up. Many open-source models now offer performance that rivals these big-name models, with the added benefit of being freely accessible.

Open-source GenAI models have become increasingly sophisticated, with robust infrastructure, embedding capabilities, and APIs that make them user-friendly. Although they may lag slightly behind the latest proprietary models, the gap is narrowing fast. This trend suggests that open-source models will continue to push the boundaries of what’s possible in GenAI.

Wrapping Up

Generative AI is an incredibly exciting and fast-moving field that's transforming how we create and interact with content. From AI generalists becoming the Swiss Army knives of the tech world to innovative companies like Reface pushing the envelope, there's so much happening.

Reface, for example, is making huge waves with its cool apps like Reface, Revive, Restyle, Letsy, InkAI, Memomet, Metahead, and Unboring. These apps show off just how versatile GenAI can be, letting users create everything from fun video animations to stylish outfit try-ons and custom tattoo designs.

Finding the right product-market fit is crucial for these apps. Adding new use cases can be a game-changer for boosting user retention. It's all about understanding your audience and keeping them engaged over the long haul.

A/B testing is another powerful tool in the GenAI toolkit, but it's important to do it right. Avoid common mistakes like overcomplicating tests or mixing up your test groups.

The GenAI landscape is evolving at lightning speed. We're seeing amazing advancements in video generation, music creation, and even multi-modal AI that combines text, audio, and visuals. Open-source models are also catching up quickly, offering powerful alternatives to proprietary systems and driving innovation.

As we look ahead, it's clear that the future of GenAI is incredibly bright. Whether you're a developer, researcher, or just a tech enthusiast, staying on top of these trends will help you make the most of what's coming. The possibilities are endless, and we're just getting started!