NOTE AS OF 10/2/23:
Given a few limitations of the ComfyUI at the moment, I can't quite path everything how I would like. There is a merge in the works (no eta) which would allow for much greater flexibility in regards to adding workflows nicely to one main workflow.
For the time being, what I currently have will work.
V2.1 UPDATE:
Changed the IPAdapter nodes to a different repository which supports more custom nodes as well as updated to include VIT-H PLUS. Larger model, but incredibly more coherent when utilizing the IPA process. Can EASILY work as a single image input LoRa alternative.
new IPA custom_nodes pack, see notes in workflow on left for path.
VIT-H PLUS utilization by default for IPAdapter (update to latest commit)
New nodes added to IPA group to allow more finetuning of image inputs processing
Can be used as a single image input LoRa alternative.
Added second Clip_Vision model loader as Revision process must utilize VIT_G model.
This workflow is mostly created for my own purposes. However, it is highly adaptive given the ability to combine nearly every aspect of it. There are a ton of logic nodes working behind the scenes to work all the switches together properly.
Control switches operate Controlnet/Revision/Ipadapter/etc, and combinations of them together.
It's all organized by groups to keep things nice and tidy. No longer requires bypassing groups or nodes to allow switches to work properly
PLEASE READ ALL THE NOTES THROUGHOUT THE WORKFLOW. ESPECIALLY THE INTRODUCTION NOTE TO THE VERY LEFT.
Thanks
Description
Changed the IPAdapter nodes to a different repository which supports more custom nodes as well as updated to include VIT-H PLUS. Larger model, but incredibly more coherent when utilizing the IPA process. Can EASILY work as a single image input LoRa alternative.
new IPA custom_nodes pack, see notes in workflow on left for path.
VIT-H PLUS utilization by default for IPAdapter (update to latest commit)
New nodes added to IPA group to allow more finetuning of image inputs processing
Can be used as a single image input LoRa alternative.
Added second Clip_Vision model loader as Revision process must utilize VIT_G model.
FAQ
Comments (5)
This looks great however, I can't find "model-ipadapter-VitH.safetensors" anywhere, the only one I can locate from your notes is simply "model.safetensors"
Yeah that's the one, I just renamed it on my local computer so that I know which model is which when I'm selecting them. So just download that model.safetensors under where I specified for Clip Vit_H:
This looks like it could be really cool. But there's something going on which seems to be stopping it doing face restoration and recolor. There's two nodes in red which lead into your Logic covered by the blanking plate (nice touch, it looks crazy in there). One is on the positive prompt for the face detailer, and the other is on the recolor KSampler. They both have noodles which terminate "somewhere" in the logic, but I've literally been using ComfyUI for about 4 days and I'm starting to get the feeling I'm about to lose a month of my life working out how to use it. Any tips on fixing this would be absolutely fantastic. Thanks!
I'll take a look at this version and see if there's outdated nodes in it. I keep changing mine up personally so it's hard to notice when things change elsewhere
Did anyone manage to make it work?
I cannot figure out what custom nodes pack contains "Evaluate Integers" node that was not found.