A party game that paints your face with AI. Here's what that actually took.

Friendslop games are having a moment, and everyone seems to want to make a viral one and earn billions of dollars. I was talking through friendslop ideas with a friend when I came up with a simple one: you drink a potion, and you see what it does to your avatar.

With AI tools, an idea like that doesn't have to sit on a backlog. You can build a simple prototype and find out if it works, maybe even make a new game every month. People have strong feelings about that, so here's what it actually took: what AI made easy, and what it didn't.

It wouldn't leave me alone I already had other projects, so at first I told myself I wouldn't work on this one.

Then it sat with me for a few days, and I realised it would be a really fun idea.

I'd originally pictured a 3D game with rendered avatars. Instead I swapped to AI portraits made from selfies, because seeing your own face change is far funnier than watching a character you've never met. I built a prototype with Claude first, just to prove the technology could do it. Once it could, I worked out the extended mechanics and built the real game.

What I want from a party game In a multiplayer game, I think three things matter most. Anyone should be able to see what's happening, pick it up fast, and leave with something to talk about.

That means clearly showing the state of the game and how the players are interacting, so anyone can jump in or out. The only things hidden are the secrets it's meant to keep.

In You Witch! the choices are simple. You pick a good or a bad ingredient, then the Drinker decides whether to drink the potion or pass it on. At the end, everyone sees what they've turned into.

I spent a lot of time tuning how long a game lasts, so that people stay interested and the payoff arrives before anyone gets bored. The ending matters most, because a portrait is something you can laugh at together and then share.

Taking it to the pub The best playtest I had was the pub after work. I got co-workers playing and watched what they enjoyed and where they got stuck, then went away and changed things.

After that I took it to my indie game dev group, for feedback from people who think about games the way I do. In a room, with people, it went really well. That turned out to be both the good news and the problem.

The problems I didn't see coming The code was rarely the hard part. The hard parts were people, cost and taste.

Getting people comfortable with their photo. I don't keep your photo. It's used to paint your portrait and then thrown away. That turned out to be surprisingly hard to get across. At first I had a tick box to accept the terms which scared people off, and a game that had been fun suddenly felt like a contract. So I swapped it for a short plain note right where you add your photo. It explains what the photo can be and that it's thrown away after painting. Adding one is up to you and the full terms are a tap away.

Filling the wait. A portrait takes a while to generate, and that's dead time at the worst possible moment, right before the payoff. I needed a distraction that was fun but had no effect on the game.

Balancing cost and quality. I experimented with models, prompts and resolutions, and made contact sheets of every ingredient to check that each one actually worked.

Being fair to everyone's face. I wanted the game to keep people's own features, across races and genders, without falling into stereotypes. A joke about a witch's nose is fun. A portrait that changes who you are isn't.

Making ugly portraits at all. Half the game is grotesque potions, and AI models are understandably wary of making someone look ugly because it can read as racism or bullying. So I had to find a way to be silly and gross without being cruel which was the right line to be pushed towards anyway.

Yes, I built this with AI assistance. You Witch! was built with a lot of help from Claude Code, and the portraits are generated by AI image models. I'd rather be upfront about that because opinions on AI vary a lot among people who make and play games.

It's what let one person take this from a pub idea to a working game with payments and accounts. Stripe, Resend and PostHog were surprisingly easy to work with because their MCP servers and guides gave Claude the context it needed. On my own that work would have taken me far longer and is not usually the fun part of programming.

The decisions were still mine. AI didn't sit in the pub watching people get confused by a tick box and it didn't decide what was funny or where the line was with someone's face. Those were the parts that made the game work.

The doomscrolling problem Making the game turned out to be easier than getting anyone to play it.

I made gameplay and promo videos and uploaded Shorts to YouTube. One got 1,200 views overnight with no advertising, which was really exciting. I kept going, but nothing else stuck and even paying a little to promote the videos didn't bring much traction.

Eventually I worked out why. In a group, the game goes really well but someone doomscrolling alone at home can't try a party game. They'd have to remember it, gather friends, and want to play it later. And I'm still the only person who has ever run a game at this stage.

So I built a solo mode. You add a photo, pick three mystery potions and get your own portrait straight away. Now someone can actually get it within a minute, share the result, and maybe remember it next time their friends are around.

Where it is now I haven't made billions of dollars. I have made a game that makes a table of people laugh at each other's faces, and I've learned that building a thing and getting it in front of people are two very different jobs. I still have plenty of ideas for where to take it.

I also got my first hater this week. An Instagram comment on my ad called the game "AI slop made with stolen assets" and asked for scammers and thieves to be kept off the timeline. The same week, a complete stranger tried solo mode and went back for a second brew giving likes and heart emojis.

That feels like an honest picture of making a game with AI right now. One person calls it slop, and another wants to see what the next potion does.

If you want to see what three mystery potions do to you, try it solo for free. If you like the result, bring some friends and make them drink something.