KREA is building the next frontier of human creativity ⚡️

CV Deep Dive

This week, we’re featuring Diego Rodriguez, the Co-Founder of KREA.

KREA is a new startup transforming human creativity. Founded by Diego Rodriguez and Victor Perez in 2022 via the highly-selective HF0 residency, it’s become one of the most popular and feature-packed AI Creative Suites today.

KREA’s real-time generation capabilities are now being used by close to 1m users, including those in the film and gaming industries - and its founders are on a mission to build the world’s most controllable AI for human creativity.

Let’s dive in ⚡️

Read time: 7 mins

Our Chat with Diego 💬

Diego! Welcome to CV AI. Tell us about yourself and your journey with KREA.

I am Diego Rodriguez and, along with Victor Perez, we founded KREA.

KREA was born out of a frustration to create.

That frustration heavily predates the company; Victor started working on generative models around six years ago. I introduced him to AI early in college, where we met. Soon after that, he got obsessed. Next thing I knew, he was working for Joel Simmons on one of the earliest generative AI applications: ArtBreeder.

A few years later, I started grad school in NYC at Cornell, where I was doing computer science, design & psychology, and machine learning. While I was there, Victor and I kept exchanging notes about his new project to interact with AI directly in the browser. Without telling him, I applied with his idea to HF0— a residency program a-la ‘YC for technical founders’.

So, HF0 called me for a final interview in Miami and I told Victor about what I had done and asked if he wanted me to stop altogether; he told me to “sell the f*ck out of this sh*t”. Less than 72 hours later, we were both in Miami - and saw each other in person after many years. We stayed there for a week, but did horribly in the final interview.

Then, to our surprise, we got an offer from HF0 to start KREA. I thought to myself, “this is probably not going to go anywhere, but I am willing to put my graduate studies on pause to see what the future holds”.

A few months later, we raised millions from undisclosed investors and role models, such as Keith Adams (founder of Facebook AI Research) and Pamela Vagata (founding team at OpenAI), who trust in us to build the creative tool of the future: a tool rooted in a perfect symbiosis between human creativity and artificial intelligence.

There are a number of tools at the intersection of AI and creativity. What makes KREA different from an end-product perspective? And how much user growth are you seeing?

Ah, the $1T question.

Did you know that when Steve Jobs was asked—by Walter Isaacson—which product he was proudest of, he didn’t reply with Mac, iPod, or even iPhone?

As a technologist living in San Francisco, I am tempted to steal Jobs’ style to answer this question—and I will do it: the difference between the products that KREA builds and others is that they were not made by the KREA team.

The beauty of the team building KREA is that the people working on it are musicians, poets, photographers, graffiti artists, and writers who also happen to be great at playing with computers and artificial intelligence.

We are getting close to a million users and we just had our biggest growth week as we’re doing this newsletter.

Walk us through KREA’s feature-set. Aside from Real Time Generation and Upscale & Enhance - what else should users experiment with?

The best way to learn how KREA works is by using it yourself; the second best way to learn how KREA works is seeing how a creative uses KREA. We see kids and grandmothers using KREA to create art—that’s our benchmark for human success.

We also allow you to train your own models on KREA but the interface is rudimentary.

What’s the most underrated feature you think KREA has? Which use-case has surprised you the most? 

I personally have underrated the Real-Time Generation, which should be almost punishable; humans should never bet against human creativity.

For example, I knew people would find a way to connect their iPads to KREA to start painting and let the AI imagine new styles and textures in real-time for their paintings. Ok? Well, I did not expect kids using KREA to render Minecraft gameplay footage or Hollywood professionals using it to decorate a movie set using projectors so that actors could get a sense of a movie scene in real-time (as opposed to 3D modeling it during the span of days or over a week).

We are only scratching the surface, KREA is still in beta.

What’s your approach to choosing what to build - e.g., how did you decide to build the QR code feature, or real-time image generation?

Following curiosity relentlessly and embracing entropy.

This manifests itself in many ways: reading papers, running experiments, talking with artists whom we love (e.g. BOLDTRON), walking, dreaming, exercising a lot, and talking with role models from industry (e.g. the other day we just had a guest from Pixar at home). Oh, and KREA dinners.

For the earliest version of KREA Realtime, we discovered it at a KREA Dinner. KREA Dinners are cozy events at our place where we bring artists, engineers, and people who we think would be great if they knew each other. For one of them, Victor built a prototype using real-time from a paper that had been released 24 hours before and we connected it to a webcam. When we saw how people were paying more attention to the camera than the guests, we thought it would be cool to ship that as a tool. 

We did a tweet showing what the KREA Realtime tool would look like (it was running on localhost). People thought it was already available and started subscribing just because they thought they’d get priority access. In less than a day, we made $10,000 in subscriptions. That’s when we thought: “okay, creatives want this”.

How do you ship so fast? And how do you keep up with the latest in the real-time / inference field when things change every day?

We are enthralled about what we do and we work indefatigably.

What are some of the major challenges you’ve faced so far? 

A previous challenge I had was learning to unfold the part of me filled with pain and realizing that “the struggle” is a necessary (yet not sufficient!) condition for building something great. I now believe in a mix of asceticism and Epictetism.

My current challenge is rediscovering how my brain works and how I can best put it to use for KREA—all without swimming against the current or flow of what’s best for humanity, that’d be a waste of my time and everyone involved at KREA.

This means prioritizing two things:

1. Finding my strengths and weaknesses. After that, I will multiply my strengths and surround myself with people who can compensate for what I am not the best in the world.

2. Balancing vital struggle and gratuitous pain. If done well, this means the difference between pushing someone to do their life’s work and exploiting them; I always practice this power struggle with myself first.

What’s the hardest technical challenge around building KREA? 

Integrating art, science, and engineering while delivering shareholder value, and fulfilling my fiduciary duty—all while being morally and ethically responsible—has to be one of the hardest things I’ve ever done in my life.

Now, from a strictly engineering point of view, I’d say… everything. It’s hard for me to pinpoint a single thing; I hold everyone who joins in an extremely high bar. From crafting polished UIs, repelling cruft at the programming interface level to accelerate shipping speed, a reliable and observable infrastructure… all the way to having excellent controllable AI models (we just started thinking about the future research team at KREA).

In summary, orchestrating all the moving pieces while ensuring fast shipping speed without killing great larval ideas—this last part is extremely important for a company that manufactures creativity.

How do you see KREA evolving over the next 6-12 months?

Building the most controllable AI for creative uses. This will involve better models at the AI research side, better reliability and speed at the engineering side, more intuitiveness at the product side, and more revenue on the financial side.

We think the sweet spot is 12 people; we are currently at 5 full-time employees—and someone just flew out of Hawaii and is moving into our house to start the interview process at the KREA house (most people work for weeks, if not months, before joining full-time at KREA; it’s not for everyone).

What’s your perspective on the debate between open-source and close-sourced when it comes to AI?

I see a future for both. However, to this day I have not seen a well-integrated product purely or mainly based on open-source software: I love Linux, but it’s honestly a mess—and the same applies for Android devices.

History doesn’t repeat itself, but it often rhymes… and open-source does not seem to rhyme with well-integrated products—which makes sense given OSS over-emphasis on decoupling.

Now, we owe many things to open-source software, research, and datasets. So I think it is a moral imperative for technology companies to give back to the community as soon as they can afford it. I still don’t know what that looks like for KREA, but I think about it at least twice per month.

How would you describe the culture at KREA? Are you hiring? What do you look for in prospective team members? 

We believe pain predates growth. We create. We care about each other. We embrace personality differences in the pursuit of a common goal. I believe “Take care of the people, the products, and the profits—in that order” is a good business mantra.

Victor and I know that the first 10 people are going to be one of the most important people for KREA. We have found 5—including ourselves. We are looking for 5 other people: mainly software engineers, designers who code, and AI researchers.

In essence, we ask for one thing: caring.

This mission is extremely important to us. That’s reflected in the amount of human years we’ve spent honing our skills, rejecting more financially rewarding or less painful opportunities, disagreeing thoughtfully, in the hundreds of pages and hours that we’ve meditated about KREA, and in the thousands of kilometers that separate our families and friends from the founding team (many of our loved ones do not live in the US).

It’s okay if you don’t care about the future of creativity and AI, but do not apply to KREA if you do not care about our life’s work.

If you care, send me an email to d at krea dot ai explaining pithily why you belong here.

What advice would you give to founders looking to pursue AI research or challenging projects in AI? 

The way I give advice is answering the question “what would you do right now if you knew what you know but your life circumstances were slightly different?”.

In this case, I’m going to assume that the slight difference is “you are not the CTO and Co-founder of KREA, a company that you are committed to”.

Well, in that case I would say…

Remember that there is no rush to get started early on a life-long pursuit. Earn enough to not die of starvation and do something uncomfortable—and please, make it count. Run under the rain for 10 miles, meditate for 2 hours, learn to do an airflare, participate in a local archery tournament, or, say, implement your own physically-based renderer using Rust and create an auto-differentiation framework from scratch using C++ understanding parallel reductions using only CUDA kernels.

All of the above are more intense variations of things I’ve done that I don’t regret doing.

The pithiest moments of my life were either chosen irrationally or I was unaware that a great choice was being made. My best friends, my best company, my best team, my best investors, or my best love were all happy little accidents. AI will be no different.

Any thoughts or predictions about what we’re going to see happen in AI in 2024? 

I expect many shallow AI products to start disappearing between this year and the next one. I also think that this year will be starred by 3D and video models. I also think that more efficient models will see the light of day this year.

Conclusion

That’s a wrap for our third Deep Dive of 2024! Follow Diego on Twitter to learn more about his work, and sign up to KREA to play with the future.

Read our past few Deep Dives below:

• 2/18: Julius is transforming computation with AI

• 2/11: How USearch Reached 500k+ Python Downloads

If you would like CV to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter.

Join Slack | All Events | Jobs | Home