My Simple To Understand Workflow
v1.0 - Initial release for Z-image
v1.1 - Some cleanup and set the model to Krea2. But can still be used with Z-image
v1.2 - Moved ModelSamplingAuraFlow out of the active workflow.
Ramble
I have noticed a lot of frustration with "simple" workflows that is far from simple. And i was also extremely frustrated by huge complex workflows that was so hard to get working it felt like the workflows was made by someone that didnt have the ability to look at things from others perspective. I hope this workflow will be give the oposite experience, and feel like a workflow you understand as soon as you open it and run it the first time.
Lora Manager
IMPORTANT: Lora Manager is a separate website from comfyui that is available when your comfyui server is running. Just add /loras behind the port in the url when you are in comfyui, or press the little "L" button to the left of your "Run" button.
I can not put into words how much i love to use Lora Manager to manage all my LoRAs. And Lora Manager is the sole reason i got hooked to LoRAs. I had over 3000 LoRAs at a point, and without LM this would have been a nightmare. But LM sort everything by model and so many more qol features.
This is also where you apply LoRAs to the Lora node in this workflow.
And personally i support the creator of it by subscribing to his crome extention witch let me download LoRAs directly into the LoRA folder from civitai. It will also add some nice features that show what LoRAs you have downloaded or not. And let you download LoRAs directly from the thumbnail. And many more freatures that makes life so much simpler.
I also recommend adding a API key from civitai to Lora Managers settings. This will let Lora Manager download all the thumbnails etc directly from civitai.
Layout
I like my workflows in a line, so that is what you get. A workflow that you can use as is, or as a template to expand to your own needs.
It's a 3 stage workflow with a main sampler, refiner sampler and a face detailer. And you can bypass the refiner or face detailer individually.
I did not put everything into tons of subgraphs, so it should make it simpler to understand what's happening if you are new to ComfyUI. And nodes are not perfectly aligned with tons of routing etc since a workflow should be flowing and in constant change imo.
There are probably some nodes named incorrectly or other small details that should be fixed, but tbh I don't really care that much about the perfectionism in a workflow.
Custom Nodes:
There should also be minimal custom nodes needed if im not mistaken. But let me know if there is some more, and i'll remove them in a V2.
Required custom nodes:
Lora Manager
rgthree
ComfyUI-Impact-Pack + Subpack
Optional custom node:
mxtoolkit
RES4LYF for bong_tangent scheduler (recommended)
Edit:
Forgot that the face detailer uses ComfyUI-Impact-Pack and its subpack. Added to the list of nodes.
Description
FAQ
Comments (5)
Your creations have inspired me to use two Lora Loaders for my own workflow. I keep one loader filled with all of my commonly used sliders available and deactivated unless I need them, and then the other is used for image specific loras. Helps me stay organized, and makes it much easier to fine tune the images with minor tweaks to the sliders.
I like your idea, think I'll do the same, thanks
At the beginning of my Comfyui days i applied all LoRAs to the model at the start of my workflow and passed that juiced up model link all the way down my workflow.
But after i started branching off at the different stages of the workflow and applying the LoRAs directly before the samplers i feel like i have much more control over whats actually beeing applied. Without full control you could all of a sudden apply the same LoRA two places in the workflow and not understanding why every image looks shit XD
Looking at stuff like this, it really shows how much the user's own brainpower matters when using AI tools.
It's so clean, neat, and intuitive.
Completely different from my own workflow, which looks like a living proof of chaos theory 🙄.
I know this sounds like total flattery, but I genuinely mean it!
I feel like others who make those insane workflows dont acturaly use their own workflows. Noone needs 6 detailers for every bodypart, 4 upscaling steps, etc. I think i have found a balance that is both fast when needed, and can be detailed when needed.
And with comfyuis dynamic memory now adays i have no problem running full 12GB models on my poor 12GB GPU, so there is really no need for a "low vram workflow".