A1111/Forge style workflow in ComfyUI for beginners that feel lost using comfyUI and miss the UI in A1111 or Forge
Version 2.0:
Version 2 is cleaner and more user friendly. (version 1 is considered deprecated)
Features:
Base text2img generation
Upscale with ESRGAN models (ultrasharp etc.)
Upscale Latent
Adetailer (Face Detailer)
ControlNet
Guide:
use the latest version of comfyui (currently 0.3.64)
use comfyui manager (github)
install comfyui custom scripts (github)
drag and drop the json file into comfyui
install missing nodes
toggle options (upscale, controlnet, ...) on the left side of the workflow (do NOT bypass or mute node manually. just use the fast groups bypasser node to toggle them)
you have to manually add controlnet preprocessors (mentioned how inside the workflow)
(optional) click on hide links button on bottom right section in comfyui for better visibility
toggle options:
hide links:
Version 1.0:
This workflow is a brief mimic of A1111 T2I workflow for new comfy users (former A1111 users) who miss options such as Hiresfix and ADetailer.
Features
A1111 prompt style (weight normalization)
Lora tag inside your prompt without using lora loader nodes
Clip Skip, RNG and ENSD options
Hand Fix (MeshGraphormer)
Hires Fix (Latent/ModelUpscale)
Face Fix (ADetailer)
Control Net
How to use
Install ComfyUI manager if you don't already have it
Load the json file or drag and drop it inside comfyUI in browser
Open ComfyUI manager and select Install missing custom nodes
You can also drag and drop preview images in ComfyUI to get the exact wrokflow and prompts
Read important notes below to fix common problems
Important Notes
Update your comfyUI and the required nodes to the latest version (if you don't, you might get a VAE error from FaceDetailer)
Enable lora auto complete in settings (comfyui custom scripts addon)
Hand Fix sometimes cannot identify hands in your picture and returns an error if the mask is set to "tight bboxes". just set it to "based on depth".
If you want to disable any section e.g. HiresFix, You should Bypass the nodes in the group and also use the on/off swith on top of it. I know it's annoying but I couldn't fix it. if you use bypass only, it cannot proceed to next steps. If you switches only, the ksamplers still do their sampling wasting time. That's why you should do both. If you know a better way for this, please leave a comment.
You might wanna also use controlnet between handfix and latent hirefix if it messes up the hands.
Description
FAQ
Comments (4)
Hi, I'm not good at making workflows (using this GUI since it's faster than A111), but do you know what nodes to create a controlnet based on openpose?
Hi, first make sure you have installed the missing nodes after loading the workflow and specially make sure you have the ComfyUI's ControlNet Auxiliary Preprocessors.
in my workflow, there are 3 controlnet preprocessors just as placeholders. You can remove them and add the "OpenPose Pose" node between "your input image" and "CR multi-controlnet stack".
just search for it or find it through: Add Node > Controlnet Preporcessors > Face and Poses Estimators > OpenPose pose (there also other preprocessors like DWpose)
@pooria1998 Hi, thanks for replying! I have the custom nodes installed and tried to use both OpenPose and DWpose as you suggested, connecting between the input image and the stack. However, mine was not following the pose after I queue it. I was testing it with a pose that raise their leg upwards, but the result keeps the leg on the ground. Alternatively, I tried using another input image with only the stickman (the one from the editor), which somehow didn't yield anything after passing the node. Other thing that I noticed from these two custom nodes are POSE_KEYPOINT, which I think it's the one that I might need to control the image for the position, I have the json file of the poses but not sure where to connect it...
Edit: Okay, I think I figured some of it. The stickman one works if I connect directly to the controlnet stack, not all but somewhat doable.
I am new to comfy (hoping to try and make ai videos) and i keep getting this error "'int' object is not subscriptable" with SaveImage right above the message