11 ms·

Launch HN: Rosebud (YC S19) – Turn game descriptions into browser games

293 points by lishali88 3y ago

Hi everyone! I'm Lisha, the founder of Rosebud AI (https://www.rosebud.ai/ https://www.rosebud.ai/). We're building a platform to help users go from description to code to game. We aim to make game creation accessible to non-technical creators, so our UI provides explanations alongside the generated code. Users have created a diverse range of games on Rosebud, including top-down RPGs, AI companions, and 3D obstacle courses, all within a few hours and sometimes minutes. Here are some examples you can play and clone (to start your own project). * Anime Jester Companion: https://play.rosebud.ai/games/ba438cc4-246e-432e-b170-4e16948cd571 https://play.rosebud.ai/games/ba438cc4-246e-432e-b170-4e1694... * Chat and Care for your Digital Puppy: https://play.rosebud.ai/games/f32a8159-7acf-4db6-a82c-70296f90bbf1 https://play.rosebud.ai/games/f32a8159-7acf-4db6-a82c-70296f... * Sphere Sync (3D game: align the sphere with the right color): https://play.rosebud.ai/games/96dfd5e1-62d4-47d8-a3e9-11038c8bb5cf https://play.rosebud.ai/games/96dfd5e1-62d4-47d8-a3e9-11038c... * Basketball: https://play.rosebud.ai/games/a0e70622-e923-4517-8c1f-728dcf0db486 https://play.rosebud.ai/games/a0e70622-e923-4517-8c1f-728dcf... * Neon Waltz Generative Art: https://play.rosebud.ai/games/e32bd12b-7cc9-4f9a-b385-42ae0b096466 https://play.rosebud.ai/games/e32bd12b-7cc9-4f9a-b385-42ae0b... * Chat with Deku from My Hero Academia: https://play.rosebud.ai/games/716fd998-aab6-4185-8375-85d9eeb2adca https://play.rosebud.ai/games/716fd998-aab6-4185-8375-85d9ee... A simple way to think about Rosebud is ChatGPT + Midjourney + Replit. ChatGPT, because we give users a chat interface for this code editor so they can describe the game they want to make and generate game code; Midjourney, because we let users generate assets inside Rosebud, 2D and 3D, to be used in their games; And Replit, because Rosebud includes a browser based code editor that lets you deploy your game instantly. Sometimes, users generate a code base from scratch via prompts. Often a simpler place to start is to modify (“clone”) an existing project on Rosebud. In both cases, we need to eventually convert user descriptions and modifications of the game into edits and changes to the codebase. To solve this problem, we had to experiment heavily with using LLM agents in production. Our agent framework tries to follow the instructions of user prompts by deciding when and whether to call upon a number of generative models (some for code generation, some for asset generation, some for character dialogue, and some for game ideas). It also must decide where to insert code snippets when it generates them. Often, a user is asking for ideas or something too vague, and our agent has to decide when to ask for feedback and clarifications. Not surprisingly, if we impose more constraints, on both the programming framework and game genres supported, our agent will perform better. However, the constraints on the types of games users can make and frameworks we want to support also constrains how flexible our platform is. Balancing these two factors, we decided to only support browser-based, JavaScript frameworks and focus on supporting AI NPCs that use LLMs themselves for dialogue and actions. This allows us to create abstractions that enable the agent to alter the codebase more successfully and guide the creator towards a more successful experience. Furthermore, we found that our beta testers are very creative with making AI character based games, and the resulting game is usually fun for players. How does Rosebud differ from Roblox, Unreal, or Unity simply adding a co-pilot? Incumbent game engines optimized their user-flow and tech stack before the advent of generative AI, and many of their user-flows are well established. We have the advantage of designing this game creation flow from the ground up. It's not just about adding code completion to an existing code editing app and including asset plugins. Such an approach wouldn't fully harness the power of LLMs. We have a chat-first interface, and having identified the limitations of agents, we can create more safeguards for users where failure is likely. Our approach will make it possible for non technical creators to also contribute to making games. Check it out for yourself! To try Rosebud: (1) head over to https://www.rosebud.ai/hn https://www.rosebud.ai/hn for access to our Discord beta tester channel and a special role. (2) then go to https://play.rosebud.ai https://play.rosebud.ai and use the code HelloHN to get immediate access. We have an array of trending projects that users can clone and mod to get started, including various character chat based games. Here’s a video onboarding of Rosebud in action: https://www.youtube.com/watch?v=h99H3FefxU0&ab_channel=RosebudAI https://www.youtube.com/watch?v=h99H3FefxU0&ab_channel=Roseb... Re business model, we plan on following in Roblox’s footsteps, i.e. keep it free for developers and take a cut of what they can charge users. Since AI tools cost more from usage than just hosting, we may have to evolve that model and see what the unit economics are (and separate a premium versus free tier for devs). (Oh and in case you’re wondering why a YC S19 startup is launching now: we basically pivoted. We were always in consumer generative AI, but focused on images until this spring, but always wanted to focus on games–Rosebud is in fact a reference to the cheat code in The Sims. When code gen got good enough this year to work for UGC in gaming, we decided the time had finally come and switched.) Some encouraging user feedback from our beta: “I have done some modding before, and I must say, this is much easier. Even when I occasionally need to code, the AI can answer all my questions and tell me how to achieve what I want. Normally, I would have to conduct numerous Google searches. What you guys have created is truly amazing.” “I’ve used Chat GPT to help me code simple games in Unity. This seems more connected and easier to work through.” “This is fascinating. This is ** amazing. Yeah, I know it's obviously early on, but already works for rad generative art. I'll say that much.” “Can finally call myself a game developer lol. Damn that sounds so good.” We’re a small team working on this for the last few months, so a lot of things are far from perfect. Constructive feedback is very welcome!

pimlottc 3y ago
Had a stumble reading the title thinking it was about turn-based games. Maybe a different verb would be better?
- amelius 3y ago
  Maybe increase the lookahead of your parser.
hubraumhugo 3y ago
Congrats on the launch! I'm excited about AI use cases in the game industry and I believe that things like text-to-asset or text-to-NPC will end up being part of every major game engine. You mentioned you've been working on a consumer gen AI product back in your YC batch 4.5 years ago, which was "pre AI-hype". Do you mind telling a bit more about your journey and pivot?
- lishali88 3y ago
  Thank you! We're so excited about how llms are expanding widely the number of people who can write software, it's the perfect opportunity to apply it to gaming since I think there's far more people who can make games if not limited to technical skills! Regarding my path with Rosebud, it's such a long journey, I'd love to share. Going way back in 2017 I just finished up a PhD in Deep Learning from Berkeley and did the less usual thing of joining an early stage venture firm. I've always had to compartmentalize my creative work from my technical before. Once the first batch of interesting results in generative came out, primarily in the image space (cycleGAN, StyleGAN …etc), I realized that it was inevitable that most creative work can be changed completely by this research. Who gets to create, how easily they can create and even what is created will change dramatically. I just had very strong conviction on this. The only thing that was hard to predict was whether this was going to happen in the next 2-10 years. Regardless of the uncertainty in when this was going happen, I knew that I wanted to get my hands dirty and build, because it’s just something I wanted to use. So I jumped into the deep-end and founded Rosebud. Earlier on, the models were quite new and not of the highest quality, so I learned quickly that consumers are a much better audience than using this for businesses. Consumers are still very picky, but also very experimental, whereas business had very specific requirements for what they want that are not easily “replaced” by AI generated stuff. Given these learnings, we iterated many times with different consumer mobile apps, which is a crucible for getting the right intuitive interface down. Long story short, you cannot go viral organically on mobile if your app is hard to use and doesn’t give a sense of magic. My core thesis is that the strength of generative AI is in its potential to make creation widely accessible, on demand and delightful. Naturally, some of our earlier apps were focused on more meme like creation experiences (like Tokkingheads, which organically grew to several million users). We learned that we must in a few clicks as possible allow the creator to achieve impressive results they want to share on social. These experiences helped build a lot of our technical ability to train and productionize models (back then there were no AI inference companies, so we had to manage everything ourselves on cloud or our own machines). Every year since the founding of Rosebud, I looked at gaming as a target application area. As mentioned, Rosebud was named after the cheat code in the Sims, which allowed me to effectively use the game as a 3D playground to build virtual worlds back when I was a kid. This was my North Star for how powerful generative AI should get, to ultimately allow users to build their immersive virtual worlds and games. I also knew that just doing game asset generation was not enough to be interesting of a platform shift for games. Was waiting on code gen to get good enough to really have an opportunity to change how game dev and creation happens.

293 points by lishali88 3y ago

pimlottc 3y ago

Had a stumble reading the title thinking it was about turn-based games. Maybe a different verb would be better?

amelius 3y ago
Maybe increase the lookahead of your parser.

hubraumhugo 3y ago

Congrats on the launch! I'm excited about AI use cases in the game industry and I believe that things like text-to-asset or text-to-NPC will end up being part of every major game engine. You mentioned you've been working on a consumer gen AI product back in your YC batch 4.5 years ago, which was "pre AI-hype". Do you mind telling a bit more about your journey and pivot?

lishali88 3y ago
Thank you! We're so excited about how llms are expanding widely the number of people who can write software, it's the perfect opportunity to apply it to gaming since I think there's far more people who can make games if not limited to technical skills! Regarding my path with Rosebud, it's such a long journey, I'd love to share. Going way back in 2017 I just finished up a PhD in Deep Learning from Berkeley and did the less usual thing of joining an early stage venture firm. I've always had to compartmentalize my creative work from my technical before. Once the first batch of interesting results in generative came out, primarily in the image space (cycleGAN, StyleGAN …etc), I realized that it was inevitable that most creative work can be changed completely by this research. Who gets to create, how easily they can create and even what is created will change dramatically. I just had very strong conviction on this. The only thing that was hard to predict was whether this was going to happen in the next 2-10 years. Regardless of the uncertainty in when this was going happen, I knew that I wanted to get my hands dirty and build, because it’s just something I wanted to use. So I jumped into the deep-end and founded Rosebud. Earlier on, the models were quite new and not of the highest quality, so I learned quickly that consumers are a much better audience than using this for businesses. Consumers are still very picky, but also very experimental, whereas business had very specific requirements for what they want that are not easily “replaced” by AI generated stuff. Given these learnings, we iterated many times with different consumer mobile apps, which is a crucible for getting the right intuitive interface down. Long story short, you cannot go viral organically on mobile if your app is hard to use and doesn’t give a sense of magic. My core thesis is that the strength of generative AI is in its potential to make creation widely accessible, on demand and delightful. Naturally, some of our earlier apps were focused on more meme like creation experiences (like Tokkingheads, which organically grew to several million users). We learned that we must in a few clicks as possible allow the creator to achieve impressive results they want to share on social. These experiences helped build a lot of our technical ability to train and productionize models (back then there were no AI inference companies, so we had to manage everything ourselves on cloud or our own machines). Every year since the founding of Rosebud, I looked at gaming as a target application area. As mentioned, Rosebud was named after the cheat code in the Sims, which allowed me to effectively use the game as a 3D playground to build virtual worlds back when I was a kid. This was my North Star for how powerful generative AI should get, to ultimately allow users to build their immersive virtual worlds and games. I also knew that just doing game asset generation was not enough to be interesting of a platform shift for games. Was waiting on code gen to get good enough to really have an opportunity to change how game dev and creation happens.