--> YUCK! THE LATEST COMFYUI UPDATE BROKE (some of) THE comfyui-various NODES <--
I am currently looking into it and see what can be done...
This workflow for ComfyUI creates a wide range of random characters in different clothing and situations. It optionally generates random sceneries without people as well. Made for SDXL / Pony / Illustrious models. The workflow comes with convenient switches for different content types:
SFW / NSFW🔥
Realism / Illustration
Male / Female
Indoors / Outdoors
Person / Scenery only
Load a model and generate a bunch of random images to test its capabilities and prompt adherence. The workflow includes an optional LoRA stack, LoRA tool**, upscaler and x/y script which gives you all tools you need to quickly test a model. Once you find an interesting prompt you can switch to manual mode and copy the random prompt to the manual prompt node.
I also added a FreeU and a Kohya Deep Shrink option wich allow you to tinker with output quality even more. You can also disable the "Pre text" and "Negative prompt" nodes.
** shows you the meta tags (i.e. trigger words) of a LoRA
Description
v2.1
- random/manual prompt switch
- x/y plot (you might run out of memory if upscaler is active)
- added a ton of tags
FAQ
Comments (7)
Doesnt work. Some issue in EasyLora Stack node
You probably have a different folder structure, so you must select a model to make sure the node can load one. If that is not the issue, please show me the console output of the error.
EDIT: oh, and maybe it's the same with the LoRA. You must select one that is actually installed on your machine.
It set for 1 LoRA, but actually have 2 of them, expand node to use 2, and then set second one to none. For some strange reason node seems to use even hidden LoRas
@NRieko yeah, that's a bug of the efficiency loader and I already filed a bug on github. The workaround is to right click on the loader node and "reload" it. After that you must set the batch size to 1 again, because the node resets to a 512 batch which leads to an OOM error.
If you find a loader that doesn't have this issue please let me know.
@NRieko oh... and I forgot to mention: make sure the Lora stack is enabled ;)
Is it possible to request one for Illustrious checkpoints? I'm not sure (only checked the workflow's image), but probably only the quality tags need to be changed.
I tested this with several Illustrious models and I could see no difference with or without the pony-specific tags. So I just left them in there. You can remove them of course - you find them in the 'Camera & Light' text node.