Built from @anshuc’s Dream Loop skill 👇
• @OpenAI GPT-6 Astra for most of the experience
• @OpenAI GPT-5.6 Sol for selected tasks to save tokens
• @MeshyAI for the character models
• @ComfyUI with audio model for sound effects
• @GeminiApp for video generation
All brought together with @threejs .
https://t.co/DMo2sH3en6
@Tim_LB@anshuc@OpenAI@MeshyAI@ComfyUI@GeminiApp For sure! I did look into existing libraries, but I wanted to build it myself this time to really understand the ins and outs. Now that I’ve learned from the process, using a library would definitely speed up the next iteration.
Built from @anshuc’s Dream Loop skill 👇
• @OpenAI GPT-6 Astra for most of the experience
• @OpenAI GPT-5.6 Sol for selected tasks to save tokens
• @MeshyAI for the character models
• @ComfyUI with audio model for sound effects
• @GeminiApp for video generation
All brought together with @threejs .
https://t.co/DMo2sH3en6
Thanks! I read your article, really interesting.
Here’s the rough breakdown:
• 1h: first pass using only GPT-6 Astra (medium) and @anshuc skill. The result was already strong enough to make me want to push it further.
• 2–3h: improved the characters and scene details with Astra (low) + GPT-5.6 Sol and meshy and blender MCP's.
• ~1h: prompt-by-prompt refinement, mostly with Sol.
Also prepared character design sheets and environment concept art to keep the art direction consistent. I’m working on a more in-depth article about the whole process.
@hi_im_ByooNg@anshuc@OpenAI@MeshyAI@ComfyUI@GeminiApp Thanks for giving it a try! The city is still a little too ambitious with memory. An optimization pass is next, so your Mac ( and mine ) won’t have to make the ultimate sacrifice 🫡 😃
Thanks! I used ComfyUI for sound generation with Stable Audio 3 Medium Base. The world was built in Blender with the help of Blender MCP and GPT-6 Astra, although most of the environment is procedurally generated in code by GPT-6. For the characters, I created character-design sheets and used Meshy for the initial 3D models. I’m currently rebuilding them to achieve better quality.
@Tim_LB@anshuc Whoa, this is some serious work. Thanks for documenting the timings, costs, limitations, and even what the data can’t prove. This is genuinely useful! 🙌
@threejs Already moved Helion City to r186. The updated Sky and better depth-aware steam integration fit the project beautifully, and the scene is still holding 60 FPS. Thanks for another great release!
https://t.co/9k1Fgv8MHg
Built from @anshuc’s Dream Loop skill 👇
• @OpenAI GPT-6 Astra for most of the experience
• @OpenAI GPT-5.6 Sol for selected tasks to save tokens
• @MeshyAI for the character models
• @ComfyUI with audio model for sound effects
• @GeminiApp for video generation
All brought together with @threejs .
https://t.co/DMo2sH3en6
@preyforge@anshuc@OpenAI@MeshyAI@ComfyUI@GeminiApp You spotted him 😅 The technician character is definitely the haunted one. I’m actually reworking the character models, hopefully the next iteration will involve fewer paranormal handoffs 👻
@beiwei50@mustafaakin Are you running it in a VM or on Linux? @mustafaakin, is the game WebGPU-only, or does it also have a WebGL fallback? WebGPU support on Chrome/Linux is still limited and depends on the GPU, drivers and display server, VMs can also lack GPU acceleration.
@wik_plus@anshuc@OpenAI@MeshyAI@ComfyUI@GeminiApp I’ll take that as a compliment 😄 Now you’ve got me thinking about making Helion City more explorable. I might skip the endless loading screens and furniture microtransactions, though.