gpt4o still sucks at generating 3d models. shame. I just asked it to make a simple low-poly star:
Was hoping its improved visual understanding would help here.
Discussion
Is it rolled out already? Thought they said over the next couple of weeks
«We are beginning to roll out GPT-4o to ChatGPT Plus and Team users»
https://openai.com/index/gpt-4o-and-more-tools-to-chatgpt-free/
Yeah found it, had to log out and back in to see it in the UI🤙
i just want ai to make models for me 😭. Its the only part i don’t want to do in gamedev
To me this is a critical step in generative AI. We ask it to reason about the world without any intuitive model of what the world is. Training should involve translations between, text, audio, images, video, 3d models, and 3d animations. That would force it to develop an intermediary understanding of the world.
yes 🤝
Why would a language model make good visuals? Diffusion models are what make good visuals.
this new model does audio, images and text all at the same time, so was hoping it would be smarter here. not sure what their architecture looks like
Surely they will eventually combine the various types of models into a single product to create a cohesive product, but maybe instead of waiting for that you should just move over to a diffusion model to make your desired imagery.
GPT is primarily a language model. OpenAI is working on diffusion models, primarily their video generation model called sora, but language models will never do imagery as good as diffusion models, so until they have something that’s combining both types of models I wouldn’t expect anything close to perfection or even “good”
its much better at generating blender scripts that generate models