Donate here if you want: https://ko-fi.com/lopi999
About
EVERYTHING HERE IS IRRELEVANT AS OF V2 > This node was something I had in idea for a long while, but I wasn't too sure how to actually implement a proper regional prompter. The most popular method from a workflow that's pretty popular on here is to use the cgem156 Attention node, unfortunately this node has quite a lot of issues. For one, it would throw out an error for no particular reason when doing hires fix. This was then "replaced" by the A8R8 Attention nodes in a recent version of that workflow... which also just didn't work for my purposes. The Impact node family has regional prompting nodes, but they all require being thrown into a specific sampler that also just didn't work half the time. Finally, there was the MultiAreaConditioning node a part of the Dave_CustomNode family, but this node is now broken with newer ComfyUI versions.
Finally, I found the FluxRegionAttention node family, and it actually seemingly just works. As thus, I decided to move forward with making a regional prompt workflow based around it. I will state this: it is not perfect. It still has some issues sometimes (eg. sometimes regions can get "swapped"), but I've also had issues with the regional prompt extensions with A1111 as well, so I don't think any regional prompting setup is 100% perfect. However, compared to other regional prompt workflows I've tried before, this is much more consistent (at least for me) than all of them. < EVERYTHING HERE IS IRRELEVANT AS OF V2
This workflow is designed by default to create a simple two region setup, with two columns. It is very easily extensible, and you can use the mask output of a load image node to create fully custom masks without needing to use the region bbox or nodes in the mask setup group.
Changelog
V2:
A8R8 either got an update or I got over my skill issues using it. Overall I was planning to revolve a regional prompt setup around it, and namely, also having an easier mask setup section as well. The workflow has been completely and utterly redesigned from the ground up, and with more options to get metadata and such.
A note
This workflow contains a lot of custom nodes. Before commenting that you have an issue with a node installing...
Did you read the list of all custom node family/families that failed to import when you restarted or installed a node family, and did you read the console log to see why?
Did you update ComfyUI and/or did "update all" in Comfy Manager?
Did you update Python to the latest version?
If all else fails, try to do a clean install of ComfyUI and try again. If it still doesn't work, then it is an issue on your end entirely and you need to diagnose yourself as to what the issue could be. Go to GitHub and open an issue for instance.
Description
FAQ
Comments (5)
Hey, can you list the links of the nodes you used?
Shouldn't you see the node suites in the install missing custom nodes tab in Comfy Manager?
@lopi9999 as you know the comfyUI manager is so stable that it is not able to install it.
@krigeta Search the node names, you should be able to find the suites they're from. Properties of a node should also list them. Comfy Manager always does the job for me when finding nodes so...
Wow, I spent a full day trying to build my own workflow to get 2 characters going, failed hard, got so frustrated. After a bit of a learning curve, got this workflow going, even set up 3 regions, now I'm consistently getting 3 characters interacting with each other and it looks so good and natural a lot of the time! Of course, the characters are all OCs whose features sometimes change, but man, that surprise is half the reason I'm doing this.
I'm still new to this whole scene, so I'm still discovering what's possible, but this really opened my eyes. It's insane how far some prompts can stretch into the other masks, and it's crazy how some somewhat vague prompts can keep showing you different interpretations. Now I just need to learn how to hook up some LoRAs to the workflow, there's a vague description on how to do it but in my current state, I'm just too dumb. Still, thank you for this.