We launched Mixup on Friday — and as of today, it’s open to everyone (no invite codes needed).
Here’s a little behind-the-scenes look at how it was made...




In Late August, we tried Nano Banana and were stunned by how well it preserved a person's likeness. I couldn't stop sending images to our slack 😆

I quickly hacked together a Mad Libs-style prototype in AI Studio

We decided it was compelling enough to explore further, so we sketched what an app might look like.
That’s where the idea of “recipes” came from — prompts with fill-in-the-blank ingredients like photos, text, or lists. Users could try others’ recipes or make their own.


A week later, we had a first version of the app with just the recipe creation screen.

The next week, we added profiles, a home feed, and a recipe feed



We added a drawer for recent photos and instantly found it saved a ton of time as compared to selecting photos as inputs in ChatGPT, Gemini, etc.

In October, we launched an Alpha to friends and family



Then, a funny thing happened. For the first time since starting Things four years ago, Bruno, Nick and my wives were using what we had built without us asking! Even my parents in their 70s loved it. My kids loved it. My sister loved it. It felt like we were on to something...



People started screenshotting outputs to reuse as inputs for new generations, so we added a Generations tab to make that easier. More fun ensued.

From there, we were full steam ahead. But at 3.9 cents per generation, we couldn't just give it away for free. So we spent a while implementing a credit system and subscription plans.

There’s also a lot of safety risk because people are… well, people. Despite Gemini having significant safety measures in place, we added our own additional AI moderation to keep things PG. We extended similar AI moderation across all user input-- comments, user bios, etc.

We also added AI where useful, like suggesting list items when you're making a recipe and a title once you've finished one.


Then we sprinkled in some playful AI moments — during generation, when you open an empty comment window, etc.


Now why do we think any of this is compelling? Three reasons:
1. The blank text box problem
When you generate images with AI, you’re usually met with a blank text box. It’s hard to think of what to write or what will work.

In contrast, in Mixup we show you a feed of posts with the prompt used underneath each one.

2,35K
16
De inhoud op deze pagina wordt geleverd door derden. Tenzij anders vermeld, is OKX niet de auteur van het (de) geciteerde artikel(en) en claimt geen auteursrecht op de materialen. De inhoud is alleen bedoeld voor informatieve doeleinden en vertegenwoordigt niet de standpunten van OKX. Het is niet bedoeld als een goedkeuring van welke aard dan ook en mag niet worden beschouwd als beleggingsadvies of een uitnodiging tot het kopen of verkopen van digitale bezittingen. Voor zover generatieve AI wordt gebruikt om samenvattingen of andere informatie te verstrekken, kan deze door AI gegenereerde inhoud onnauwkeurig of inconsistent zijn. Lees het gelinkte artikel voor meer details en informatie. OKX is niet verantwoordelijk voor inhoud gehost op sites van een derde partij. Het bezitten van digitale activa, waaronder stablecoins en NFT's, brengt een hoge mate van risico met zich mee en de waarde van deze activa kan sterk fluctueren. Overweeg zorgvuldig of de handel in of het bezit van digitale activa geschikt voor je is in het licht van je financiële situatie.

