i have a how-to article for this version! find it here.
welcome to my workflow! version 2.5!
i want to thank everyone who has appreciated my images, which were generated using some version of this workflow, and everyone who has downloaded this and earlier versions. please leave a comment if you have any questions, i am happy to help!
if you have any trouble with finding or placing some of the weirder nodes and files, or you're getting errors that are super unhelpful, please please let me know! i am happy to assist if i can.
version 2.6 - new upscaler and detailer from lonecat! reverted the clip encoder because it started getting crashy. separated the detailers into a new area 3 and renumbered 3-5 as 4-6. updated the observer area to be less fancy.
what is this workflow for?
this workflow is for human and human-like character-focused images. it's designed around controlnet preprocessors that can pull randomly from images (poses) you have collected and provides lots and lots and lots of fine-tuning and options. the goal is to make it possible to set up a basic prompt and then generate a bunch of images using the pose templates you have organized.
how is this workflow organized?
there are six bookmarked areas (1 through 6). when you're first getting used to the workflow, go through each bookmarked area one by one and get things set up. later, most of your attention will be spent on bookmarks 4, 5 and 6. area 6 is where you watch everything come together when you start a run.
thanks again!
if you have any questions i will do my best to answer them. this was my first real workflow and it's probably got some issues and it has way too many noodles running all over. but i think it works pretty well and i hope it's easy to understand. thank you!
--sft--
Description
version 1.7 - this version adds the ability to use a checkpoint model! also i moved the second sampler controlnet toggle to area 4, which makes more sense, and moved the face inpainting settings into area 3 so they're easier to find. fixed a few rarely encountered errors introduced in previous versions.
FAQ
Comments (4)
The workflow keep on giving this error:
AILab_QwenVL_Advanced
Using bitsandbytes 8-bit quantization requires the latest version of bitsandbytes: pip install -U bitsandbytes
Any help would be greatly appreciated
are you running the comfyui beta? I think the qwenvl stuff might still require it. you could also switch to fp16 and see if that fixes it. there are two places to try to switch. both of them are below area 5 (the prompt group), labeled qwenvl pose description and qwenvl prompt expander. both of them are subgraphs but if you go inside the subgraphs you will see the qwenvl node to fix. try switching the quantization setting to none (fp16). let me know if that fixes it!
i double checked and i'm wrong about the beta, it's not necessary for qwenvl, it's for the controlnet. so definitely try switching to fp16 in the two qwenvl nodes.
Come on just ask GPT or Grok. And the error literally gives you the pip command to run, so use it 👌