GenieLabs

Can AI fully automate playable ad creation? I didn't think so. So I tried the hardest version: one shot, no follow-ups. Playables are hard. A tiny game, a sales pitch in 20 seconds, in one HTML file. Studios take days on these. I picked Cut the Rope, one of my all-time favorite games. I gave OpenAI GPT Astra 6 the game's APK and one prompt. This is what came back. Why it should work: a playable is mostly code and structure, which is exactly what LLMs are good at. The APK gives it the assets and the mechanic. The prompt gives it the goal. The model reads a real game and writes a small one. Why it shouldn't: the parts that sell a playable aren't code. The hook is a feeling in the first two seconds. The brand is a hundred small choices an art director makes without thinking. There's no file in the APK for those. My score, honest: Gameplay 5/5. Cut the rope, physics, three levels. It just works. Narrative 4/5. She's freezing, light the fire. Clear stakes in one line. Creative 3/5. Solid. Not surprising. Hook 2/5. It opens like a game, not like an ad. Nobody stops scrolling. Brand 2/5. The character is there. The feel isn't. So: automate, no. Skip the first two days of work, yes. What would you have scored it?

Can AI fully automate Spine 2D character animation? We don't know yet. That's why we're working on it. We built a Claude skill that takes separated character PNGs and outputs a rigged Spine skeleton with animations — no manual bone placement. It works for humanoids. It breaks on stylized art. Motion understanding is basically unsolved. Open sourced everything while we figure it out. If this is your kind of problem — come contribute, or just follow along. Would love more eyes on it. 🔗 https://lnkd.in/d98SPsP8

Last week my playable scored 2/5 on the hook. So I gave Claude Opus 5.5 one more prompt. This is the whole thing: "This playable is a playable I made but it isn't hooky enough. I want to create 3 other variations that will drive different emotional triggers from the user side: Ego Correction (I can solve it better), Fight or Flight, What Happens Next." Same game, same girl, same fire. Three reasons to stop scrolling. Hooks aren't a design problem. They're psychology. People stop because they think they can do better, because time is running out, or because they need to know what happens next. Marketers have known this forever. The model has read all of it. So I named the emotion and let it work. Ego: someone else botched the cut. Can you do better? Urgency: 9 seconds on the clock. She's freezing. Curiosity: something is knocking at the window. What's in the box? I asked for three hooks. I got three ads. Fight or Flight ends with her frozen. Curiosity ends with a second knock. The model took the emotion as a brief for the whole thing, not the first frame. What it can't tell me is which one wins. That's not a prompt. That's a test. Which one would you have tapped?

Every vendor wants to sell you more concepts. Ask your team what they are waiting on. It is rarely the idea. It is the build, the designer, the export, the second round. Genie Playable gives the creative manager the build: start from your game, change what you want, adapt for every network, without pulling a developer off the game.

Swapping colours on a template is not a new playable. Here is what changes between versions when the source is your game, not a template. On a demo game, slide by slide.

The match-3 ad everyone is running works for a reason. The problem is it's not your game. Genie Playable takes a format that is winning right now and rebuilds it with your characters, your art, your brand. Not a reskin with your logo on it. Then you edit it, make the variations and ship to every network.