they have open source datasets like LAION-5B, the Common Crawl, the Pile etc.. the main thing is getting a model trained on it. I guess I would need to design the transformer architecture, train the model, then use that as a proof-of-concept to get investors interested.

Or build an app around Stable-Diffusion, post it on app store, make some money, and use that to get investors interested. Probably do this then with the investments and some hired help do what i said in the first paragraph