An AI Image Company Bought an Astrology App. The Real Buy Was the Persuasion Layer.

On July 23, Midjourney, the AI image-generation company that turns text prompts into pictures, announced it had bought Co-Star, the astrology app. Per Gizmodo citing Bloomberg, the deal actually closed in spring 2026; founder David Holz disclosed it later. Most wire coverage treated the deal as a curiosity. The deeper read is that Midjourney did not buy a horoscope product at all.

It bought a product organization.

Four years into a Discord-only product, Midjourney has 21 million members on the platform's largest server, no free tier, no iPhone or Android app, and roughly $500 million in annualized revenue. The company took zero outside investment, turned profitable inside six months of its 2022 launch, and runs on a single distribution surface. Estimated revenue climbed from $50 million in 2022 to $200 million in 2023, $300 million in 2024, and $500 million annualized in 2026. Those numbers are labeled as estimates by oztalking, and Midjourney confirms neither. Headcount estimates range from 11 to 163 across outlets; Forbes's April 2026 profile put the company at 60.

Co-Star is the layer Midjourney has never built. The astrology app did not win by reading horoscopes better than every competitor. It won by designing the moment after the reading: a daily push notification, a pull-to-refresh ritual, a share card that turns a one-line forecast into something the user sends to a friend or screenshots for later. Those mechanics convert model output into behavior. As oztalking frames it, Co-Star is "a product-and-design organization built to make people apply unverifiable outputs to their own lives." Holz wrote that he met Co-Star founder Banu Guler "a little over 5 years ago as I was thinking about the ways people reflect and make decisions." Co-Star spent a decade building the product surface where a paragraph a model cannot verify gets applied to a relationship, a career move, a Tuesday morning.

A $500M-revenue image generator has never built that layer. Image quality ships from Midjourney's own lab. Whether the picture a user generates gets opened again tomorrow is product-design work, and the company has staffed that work thin.

Holz has said roughly half a dozen similarly ambitious projects will surface in the next six months. Midjourney Medical, a healthcare imaging effort, is the named first signpost of a new direction the company is framing internally as a "research lab." Bloomberg, per Gizmodo, reports Midjourney plans an astrology-focused image generator as the company's first product outside Discord.

Google's Nano Banana is, per Gizmodo's reporting, already considered "head and shoulders above Midjourney" on raw image generation. None of those bets will win on model quality alone, because the quality bar is now set by teams with more compute than Midjourney can spend on any single model.

The moat that survives that is product. Co-Star built the surface for text. Midjourney now has a path to build it for images. The Co-Star playbook (daily ritual, push-driven return, shareable card) has never been applied to AI-generated images at the consumer scale. Holz has not said what the first product using the Co-Star team will look like. He has said the half-dozen projects start landing in the next six months.