Update: The 1.1 version swaps out the older dynamic prompts wildcard node for a replacement that's a full (or at least better) implementation of dynamic prompts' wildcard format. It also adds a final upscale and sharpening step that blows the image up to 4k and sharpens it with Photon v1, giving it razor sharp detail and more saturated color without messing with the image in other ways. If you don't want the second upscale step, just bypass those nodes in ComfyUI.

This workflow is great at making fantastic worlds. It works by randomly selecting three images from a folder that you specify, then combining them all with IP adapter along with your prompt, so that you end up with a result roughly analogous to an artist using an inspiration board. I've personally gathered a large folder of images and I just let it run for a while and see what it comes up with.

One caveat is that there isn't a way that I know of to get an automatic count of images in a folder, so the three random number generators need to be manually set to a maximum of one less than the number of images in your image folder (the count starts at zero). I may release a modified version of the image directory loader node with a random select setting at some point to fix this problem. Given that some things need to be tweaked, it's probably best for this not to be your first foray into ComfyUI.

This workflow uses a number of custom nodes. Use Manager to download the missing ones.

The zip file contains an A and B version of the workflow. Version A includes the IP Adapter images in the hires fix, and B does not. A's images tend to feel a bit more cartoonish and fantastical, while B's are a bit sharper and tend more toward hyperrealism.

Note: Not included in this archive are my wildcard files or IP Adapter images. I'd like to encourage everyone to come up with their own voice with this workflow. :)