https://en.m.wikipedia.org/wiki/Stable_Diffusion

This has the most basic guide to what is happening if you wish to crawl out from under that rock. It includes a built in SFW text to image prompt:

https://stable-diffusion-art.com/beginners-guide/

All one has to do is look at the NSFW images marked as being from AI and note the watermark to find where to generate images. Once you make a few, you’ll see small problems that are present in many other image categories. The main issues have to do with excluding certain prompt key words to make the output look real, then stuff like genitalia is not easy to get dialed in well unless you are running the software on your own hardware. This requires a powerful video card to generate the images and a lot of storage space. Once you know this a lot of images become obviously AI generated. There are aspects of lighting, eyes, fingers and toes, easy lighting text prompts and other small details that are harder to avoid in the image output. These start to stand out more once you know.

This tech is moving very fast right now. The next iteration of Stable Diffusion is set to release this month and it will likely make it impossible to tell what is real and what is fake. Right now SD must start with a low res image, then it can be scaled higher. SDXL will be able to start with a high res image and modify details which has not been possible. With a bit of effort, it will be possible to modify video frame by frame and use a simple text prompt to alter details. I doubt people will do more than clips at first, but with some good scripting using Blender, I could see it working for larger projects.

Follow the second posted link. And read it. This is FOSS. Combine this with an open source text to text LLVM running on native hardware and you have a real game changing set of technology.

https://generativeai.pub/how-to-setup-and-run-privategpt-a-step-by-step-guide-ab6a1544803e