1 Year of Service
The basics (after install) are as easy as the online generator. At least for ComfyUI, the tricky part was just finding (and then combining) a working workflow. (You can load other people’s workflows.) After that, the amount of effort seems to be "as much or little as you want".Been tempted to try and get into it but seems like a hell of a rabbit hole with a lot of work and a steep learning curve
Choosing models/LoRas etc is as easy as putting them in a folder and then choosing them in a drop-down menu, though you can spend as long as you want switching between the literally thousands of them that exist. (Or just use the ones recommended in the AI beginners thread.) With that, you are already more powerful than Pretty-AI.
Personally, I have no GPU whatsoever and only a cheap CPU. I’m not sure if I’d recommend that to anyone for reasons other than "wants to try it out before deciding whether to upgrade the system for it" or "want to get one specific image that is beyond the free online generators and is willing to have the computer run in the background for hours to get it right." (Single pictures take me minutes, and most are not that good.)Wouldn't mind using a local as less restrictions but does't it take a set up with crazy horsepower to run?
Based on what I’ve read, if you use a model+LoRas that fit in your VRAM, you should be able to generate at least one image per second with SD 1.5 (requires ~3GB), probably an image every few seconds with SDXL/Pony (~6.5 GB for model + a few MB for each LoRa.), or ~2 minutes per images with the faster FLUX models (any size from ~3GB to 25GB exists, quality varies.)
Given how much more powerful local gen can be, I’d definitely recommend it.