Anyone tried self-hosting 3D object generation?

Found this post: of someone fully developing a game through AI writing code, generating 3D models, etc. (he is using Meshy to create the 3D models). I was curious about creating something similar with Hermes Agent (or other?) that would develop very simple games end-to-end to benchmark different models. I tried Meshy just out of curiosity and was really impressed by what it created on the first try based on an image I made with GPT. I was curious if anyone has some experience with this? Are there any good open-source models to do this? Meshy output on the 1st try Source Image to create the 3d model. Generated by chatGPT in this instance I have very limited hardware at the moment (I use a repurposed Precision 5570 with 8GB VRAM and 32GB RAM) to run Qwen 3.8:27B through Hermes Agent. It is very slow as it runs mostly on the CPU, but I don't mind since this is more of a lab for testing and experimenting. I only run it for background tasks, never real-time usage, and I was considering if it would be possible to make Hermes develop a very simple game in the background to learn more about AI and use as a future benchmark to compare models. I was thinking about a setup like this: Hermes pilots Qwen, which does the coding and prompting. When it needs an asset, it writes a prompt, which offloads Qwen from the VRAM/RAM. Then the image/3D model generation model loads up, generates the asset, unloads, and gives the output back to Qwen, which analyzes it and iterates with vision Anyone got some experience with this or knows some good models for this? Disclaimer: I used ai to correct my post