Bootstrapped Icons8 into millions of users
with Ivan Braun, Icons8
Bootstrapped Icons8 into millions of users
Show Notes
Ivan Braun is the founder of Icons8, a bootstrapped design assets platform that started as a custom icon shop in 2012 and has grown into a catalog of over 1.4 million assets - icons, illustrations, 3D models, photos, and AI-generated imagery. He built the company over more than a decade without external funding, navigating every major design era from skeuomorphism to flat design to generative AI.
Ivan is a UX designer by training who recognized early that the only offering clients actually wanted from his agency was icons - so he leaned in, built a store, experimented with subscription pricing before it was obvious, and eventually trained generative AI models on a proprietary dataset of 80,000 consistently-lit photos of people. That project became Generated Photos, launched a full year before the generative AI wave went mainstream. He runs the company from Cariló, Argentina, where he also hosts founders and digital nomads at a guest house he built specifically to help people learn faster and ship better.
How Icons8 Started: Following the Signal, Not the Plan
Ivan was a web designer running a UX agency. The single-person web designer was a real profession then - branding, interface design, iconography all rolled into one. He created icons as a small part of client work, mostly to close deals: offer a package and say “we'll include the icons, no extra charge.”
What he noticed: clients ignored every other offering. They wanted the icons. So he built a single-page website with 40 icons, priced at $40 each, and found buyers. The product defined itself by what the market pulled from him - not by what he originally planned to build.
The first real distribution spike came from self-promotion, not paid marketing. Windows Vista shipped with complex 3D-style icons that were difficult to draw. Ivan's team created a polished icon pack, posted it for free, and it spread everywhere - design communities, blogs, aggregators. A huge wave of downloads followed. Within that wave came the insight that changed the business: one downloader said he liked the icons but hated the link requirement. “How about a hundred bucks?” That transaction became the core commercial model, and they still run a version of it today.
The Subscription Pivot: The Best Pricing Advice Ivan Ever Got
Icons8 started selling icon packs as one-off purchases - the standard model for stock assets at the time. A junior marketing hire named Annie, fresh out of college, pushed Ivan to switch to subscriptions. He resisted: “I'm not sure anybody will pay $20 a month.” Her response: “They will. Charge even more.”
That was the best pricing advice Ivan says he ever received. The move to recurring revenue changed the trajectory of the business. He credits Annie specifically and notes she left for a job in the UAE shortly after - the classic founder's story of advice arriving from unexpected places and then the advisor disappearing before they can see the impact.
The lesson for founders: the person who understands what customers will pay is not always the CEO. It's often someone watching the customers from a closer distance. Listen to them, especially when they contradict your instincts about pricing.
The Link License: SEO as a Business Model
Icons8's growth engine is structural: free access to icons in exchange for a link back to the site. Hundreds of thousands of designers and developers use Icons8 assets for free, and each usage creates a backlink. Those backlinks feed Google rankings. When someone searches for icons, Icons8 appears. The people who don't want to link - commercial software products, companies building on the assets at scale - pay to remove the attribution requirement.
This model is elegant because it self-selects the paying customers. Commercial users who can't afford a visible attribution link are exactly the users with a budget to pay for removal. Ivan ran extensive paid advertising experiments - AdWords, influencer marketing, various paid channels - and none of them worked. The link license generates acquisition that paid media can't replicate, because the competitor (Shutterstock, for example) has higher average revenue per customer and can outbid Icons8 on any keyword where they both compete. Ivan can't win that auction. He doesn't have to.
Generated Photos: Building a Generative AI Company Before the Wave
Before Midjourney, before stable diffusion went mainstream, Ivan had a dataset. In attempting to solve the limits of stock photography - most stock images are 90% right for any given use case, never exactly right - his team photographed over 80,000 people under identical conditions: same lighting, same camera, same backdrop. They removed backgrounds and composited subjects into designed scenes. The compositing product failed to gain traction. The dataset was too good to abandon.
Ivan installed Nvidia's StyleGAN library - the same technology behind the then-viral “This Person Does Not Exist” website - and fine-tuned it on the consistent, high-quality photo dataset. Because the training data was so tightly controlled, the output quality was exceptional. The results were almost too good: early feedback from users was that all the generated faces were beautiful. Ivan's team added a “not beautified” mode and expanded diversity in subsequent versions.
Generated Photos was incorporated as a separate entity for two reasons: legal isolation (no prior case law existed for AI-generated face licensing) and startup eligibility for GPU cloud credits from providers like Nvidia, who offer favorable terms to startups and are less generous with established companies. This structural choice - spinning out the AI product to preserve optionality and access programs - is replicable for founders navigating the same territory.
Managing 1.4 Million Assets: From Zip Files to AI Tagging
When Icons8 had a thousand icons, Ivan spent 28 minutes looking for the right four for a mockup. He timed it. That experience made it obvious the product was not a library - it was a search problem. The company built a database, search infrastructure, and a tagging system. Title alone was not enough: a search icon might be called “search,” “magnifier,” or “magnifying glass” depending on who drew it and when. Everything needed synonyms, categories, and contextual tags.
Icons8 used Amazon Mechanical Turk for tagging at scale. As AI improved, they shifted work to automated tagging models. Today the internal tagging team is four people. The principle Ivan runs with: keep the internal team small, outsource what can be systematized, use AI for everything that AI can do reliably. The tagging infrastructure is now a competitive advantage - 1.4 million assets are only useful if users can find what they need in under a minute.
Going Global From Argentina: How to Build a US Product Without a US Address
Icons8 gets roughly a third of its revenue from the US, a third from Europe, and a third from the rest of the world. For an Argentina-based company this is counterintuitive - most non-US founders treat US market penetration as the hard problem. Ivan's framework for building an internationally viable product from anywhere:
- Use English throughout the product, documentation, and marketing. US customers, and international customers researching products, default to English.
- Use US infrastructure: servers hosted in US regions, Stripe for billing, AWS or equivalent US cloud providers.
- Don't design for any country specifically - design for international users. The US market is just the international market with larger budgets.
- From outside the US, everything you consume about business, finance, programming, and entrepreneurship is US-flavored anyway. That education translates into product instincts. Use it.
The Firing Framework: Do It Now, Do It Cleanly
Ivan's advice on letting someone go is blunt: do it now. The day you fire someone is the worst day of your week. The day after is the best day of your week. If you have already tried performance reviews, coaching, and task reassignment and none of it worked, you already know the answer. You're delaying because the conversation is uncomfortable, not because there's still a path forward.
His tactical suggestion: consider sending an email first rather than opening with the in-person conversation. The email lets the person process the initial shock privately, so the follow-up conversation - which you still need to have - happens when both people are past the first reaction. Ivan acknowledges he doesn't always do it himself, but frames it as the more humane approach.
After the decision is made: close all accounts immediately. This is uncomfortable but necessary. In the best case, a departing employee who retains access talks to colleagues and spreads negativity. In the worst case, they download customer lists, export data, or do something irreversible. Revoke access the same day.
Tools & Resources Mentioned
- Icons8.com - 1.4M+ design assets: icons, illustrations, 3D models, photos, music, and AI-generated imagery. Free with attribution link; commercial license available.
- Generated Photos - AI-generated stock photos of people, trained on Icons8's proprietary 80,000-image dataset. Includes diversity controls and a “not beautified” mode.
- Nvidia StyleGAN - The generative adversarial network library Ivan used to train Generated Photos. The same underlying technology as the “This Person Does Not Exist” website.
- Product Hunt - Ivan is among the all-time top product launchers on the platform. His launch strategy: use your email list (Icons8 has 4 million subscribers) to drive day-one votes within Product Hunt's terms, and capture the second wave of press, investors, and bloggers who cover top launches the following day.
- High Output Management by Andy Grove - Ivan's top-recommended management book. Andy Grove was CEO of Intel for decades and one of Ivan's most admired founders. Ivan says he has re-read it multiple times and it remains in his top three Kindle books.
Frameworks
Follow the Signal, Not the Plan
Ivan never set out to build an icon company. He was a web designer. The market pulled one specific product from him - icons - while ignoring everything else he offered. The business that existed in his plan (UX agency, custom development) was different from the business the market was willing to pay for. Founders who watch what customers actually want, rather than what they intended to sell, find the product faster.
Link License as Distribution Engine
Icons8's core growth mechanism: free asset access in exchange for attribution backlinks. This generates SEO-compounding distribution that paid advertising cannot match at the same unit economics. It also self-selects paying customers - the users who cannot afford visible attribution (commercial products, SaaS companies) are exactly the users with a budget to pay for license removal. Built-in distribution that improves with scale.
Proprietary Dataset as Moat
Before generative AI was mainstream, Icons8 assembled a controlled dataset of 80,000 consistently-photographed people. When generative model technology (StyleGAN) became available, they had training data no competitor could replicate quickly. The dataset quality - consistent lighting, professional retouching, same backdrop - meant the fine-tuned model outperformed generic training. The moat was built years before the opportunity became visible.
Startup Spinout for Legal and Structural Optionality
When entering genuinely new legal or technical territory, incorporating a separate entity for the new product preserves the parent company from liability and opens access to startup-only programs (cloud credits, accelerators, preferential pricing). Icons8 spun out Generated Photos specifically to access Nvidia GPU credits that a 12-year-old company would not qualify for. The structural choice was as important as the technical one.
The Firing Clock
Once you have exhausted coaching, performance reviews, and task reassignment, the firing decision is already made - you are just delaying it. The day you fire someone is the worst day of the week; the next day is the best. The longer the delay, the more organizational cost accrues without the problem resolving. Close all accounts the same day. The email-first approach (letting the person process shock before the in-person confirmation) is worth considering for the person's dignity.
FAQ
How did Icons8 grow without paid advertising?
The link license created compounding SEO distribution - every designer or developer who used free Icons8 assets linked back to the site, generating backlinks that drove organic search rankings. Ivan ran extensive paid ad experiments (AdWords, influencer marketing, display) and none of them worked. The structural reason: Icons8 has lower average revenue per customer than competitors like Shutterstock, so it loses any keyword auction where both companies bid. The link model generates acquisition that paid media cannot replicate at comparable unit economics.
How do you manage and make discoverable 1.4 million assets?
With a tagging system that goes well beyond titles. Icons need synonyms (search, magnifier, magnifying glass, find) plus category and contextual tags. Icons8 built this infrastructure starting when they had 1,000 icons and Ivan spent 28 minutes looking for four of them. They scaled first with Amazon Mechanical Turk and now primarily with AI tagging models, reducing the tagging team to four people. Search quality is a competitive advantage - a library is only as good as the ability of users to find what they need.
Why did Icons8 create Generated Photos as a separate company?
Two reasons. First, legal: when Generated Photos launched, there was no established legal framework for AI-generated likeness licensing. Isolating the product in a separate entity protected Icons8 from any downstream liability. Second, structural: GPU providers and cloud platforms offer favorable credits and programs for startups. A 12-year-old bootstrapped company doesn't qualify. The spinout made Generated Photos eligible for Nvidia credits and other startup programs that materially reduced compute costs during training.
What was the key insight behind Generated Photos?
Most stock photography is 90% right for any given use case - the lighting is wrong, the person doesn't match, the background doesn't fit. Icons8 set out to make fully composable photography by shooting 80,000 people under identical controlled conditions (same lighting, camera, backdrop) and digitally placing them in designed environments. The compositing product failed. But the dataset - 80,000 consistently-shot people - was exceptional training data for StyleGAN, Nvidia's generative model. The asset they built for one product became the foundation of a completely different one.
How should a non-US founder build a product that succeeds in the US market?
Ivan's framework: use English throughout, host infrastructure in the US (servers, billing via Stripe, US cloud regions), and design for an international user rather than a local one. The US market is, for most international founders, just the international market with larger purchasing power. Ivan notes that founders outside the US already consume business, finance, and programming content in English - that fluency translates into product instincts that work for US buyers. Don't think of it as penetrating a foreign market. Think of it as building for the world's default language.
What is Ivan's advice on hiring developers as a non-technical founder?
Don't hire individual freelancers for long-term work. A freelance developer is energetic for the first three months, then declines, then leaves for a full-time job six to nine months in. Developers need a company - colleagues to exchange ideas with, a team structure, community. Ivan's hiring constraint when he started: he had money for three developers for three months. The deadline forced him to find a revenue-generating project fast. That constraint - a hard deadline tied to runway - focused the product decision in a way that open-ended searching would not have.
What is Ivan's college advice?
Computer science at Stanford - specifically because the education he needed would have been compressed into four years rather than accumulated across 20 years of practical experience. His framing: the value isn't the credential, it's the 20 years of things you otherwise have to learn the hard way, being given to you in structured form with the people and environment to absorb them properly. The opportunity cost of not having it is measured in decades.