I don't think you're understanding what they're trying to say.
You don't need to use Grok/Claude/Copilot/Gemini/ChatGPT/etc. at all. Hugging Face currently lists 111,953 text-to-image and 2,893 text-to-video models available.
You can run an image or video generator model at home on your own hardware you currently have. This technology doesn't require a $10k rig to run. The massive data centers are only necessary for the Software as a Service providers like ChatGPT to operate their systems as fast as possible to tens of thousands of users, with dozens of agents processing multiple steps and reasoning to accomplish a task. One person making one image doesn't need much.
While the SaaS options definitely need to get better at detecting and blocking usage like CSAM... those services are by no means required. You can do this on any gaming system made in the last 5 years easily, and older if you're willing to wait longer for the result.