Before Midjourney, NightCafe was already profitable

Before Midjourney became a household name, an Australian couple was already building an AI art empire from their living room. Five years later, NightCafe is still around, completely bootstrapped, and pulling in millions.

In a recent interview with TechCrunch, co-founder Elle Russell pulled back the curtain on the platform's surprisingly robust business. Run by a tiny team of just nine people based in Cairns, Australia, NightCafe generates $4 million in annualized revenue with a 50% gross margin, translating to roughly $2 million in annual net profit. It boasts over 25 million registered users who have created nearly a billion images, alongside roughly one million monthly active visitors and 20,000 paid subscribers.

From failed wall decor to a goldmine

The company started by accident in 2019 when co-founder Angus Russell moved into a new Sydney home and could not find wall prints he liked. He set up a website where people could generate AI art and order physical prints, naming it after Vincent van Gogh's "The Night Café." The wall art sales flopped, but users loved creating images for free. When a surprise cloud hosting bill forced Angus to put up a credit paywall to stop abuse, users flooded his inbox asking to buy more credits. The business broke even virtually overnight.

The massive breakthrough came in mid-2021. When OpenAI previewed DALL-E without releasing it, open-source researchers reverse-engineered the approach with VQGAN and CLIP. Angus immediately bought hundreds of GPUs to host the open-source pipeline on NightCafe. Images from the site went viral on Reddit, yielding $17,000 in a single day and prompting Angus to quit his software engineering job at Atlassian.

The model aggregator play

Today, NightCafe does not pretend to be a deep-tech lab. In fact, it employs zero internal machine learning researchers. Instead, it operates as a slick UI wrapper and social hub layered on top of external models, including Stable Diffusion, Ideogram, Google, and Black Forest Labs' Flux.

Users pay between $4.79 and $50 per month, which undercuts standalone subscriptions like Midjourney. In return, they get model fine-tuning tools, access to daily art challenges, creator tipping, and physical merch printing. By avoiding model training, NightCafe has neatly sidestepped the massive copyright lawsuits currently plaguing foundational model creators like Midjourney and Stability AI, sheltering under platform safe-harbor protections while pushing infringement liability onto its end users.

NightCafe has also taken a much more permissive stance on controversial imagery. While Midjourney banned image generation of political figures like Donald Trump and Kamala Harris ahead of the US presidential election, NightCafe still allows it, opting merely to restrict such generations from appearing on public community feeds.

Why it matters

NightCafe is a masterclass in AI pragmatism. While venture-backed labs burn hundreds of millions to train frontier models, a nine-person team built an intensely profitable business simply by being the friendliest frontend aggregator with an active community. For casual creators and hobbyists, it remains one of the most accessible and cost-effective ways to sample different image models under one subscription without juggling separate platforms.

However, we think this middleman strategy sits on thin ice. Aggregators are fundamentally at the mercy of their upstream suppliers. If model makers polish their own consumer UIs or hike API fees, NightCafe's value proposition quickly erodes.

Furthermore, its hands-off moderation looks increasingly precarious. Leaving copyright liability to everyday users while permitting political deepfakes might goose user numbers today, but as global regulators tighten content governance, relying on Section 230 style defenses could prove to be a dangerous gamble.