馃憤 *** Like this workflow if it helped you generate amazing results! ***馃憤
馃憫 COMFYUI WORKFLOW, Simple flux generation for my loras 鈿欙笍
-NEW REVISION V1.2s.
Added WaveSpeed (Apply First Block Cache). For the installation refer to:
https://github.com/chengzeyi/Comfy-WaveSpeed
-NEW REVISION V1.0s, Simplified workflow without ezXY nodes.
***Notes***
In order for this to work with ComfyUI_ezXY repro from github, a file should be modified and comfyui restarted after.
Modify the file config.yaml for line 4 and 7. Where you see "True", put "False"
I'll create a new version of the workflow to remove nodes from that repro when I come back home. Some also wanted me to remove the top 2 groups that is used to save the image with the lora name on the image in a different folder, if you disagree with that removal, please post a comment.
***********************
After some request, asking me how I can generate those quality image from lora:
Here is the answer, a simplified workflow to generate image using flux.1 Dev model along with my loras.
You can use it as is, by only modifying what are in the red box.
1- The lora to use (some of the prompt are already set for a person, adjust if it's not your case)
2- The prompt context to add, as the background, cloth and such
To refine the result, you can play around with blue box.
De default setting should be good for most of you.
If you get a out of memory, or an, allocation on device, error, change the 1.4 latent upscale to 1.25 or bypass that node. But if you're computer can handle it, you'll see, it worth it.
When you're happy with the result you can upscale. To do so, simply activate the last group.
That group also divide the image by tile to show you how the UltimateSDupscale node will process the image. The best is to have the face entirely in 1 tile, but other tile should all get the context of the person it upscale or you'll get strange thing happening. For a upper body shot, a 2 rows with 3 cols work great. If you get strange result, put that at 2 cols.
The Controlnet Tile group is there only if you're result giving too much deformation. If you activate it, you can rise the desnoise of the ultimateSDupscale to around 0.5.
But those 2 groups are optional as, you'll see, the first result is already amazing.
The Workflow also save results at a specific places:
1- The normal result will got in (every day will be a new folder, so you will not end with a very high quantity image inside 1 folder resulting in an very low latency access to that folder): ...output/FLUX/Normal/YYYY-MM-DD
2- It will also output in an other folder if the 2 groups on top are on: ...output/FLUX/lora_test/YYYY-MM-DD
3- And if you want, if the group for ultimateSDupscale is on, in: ...output/FLUX/UltimateSDflux/YYYY-MM-DD
Description
v1.2s -Replaced nodes from ezXY in Modify Prompt group, thanks to Rich245 to have pointed this.
-Added "Apply First Block cache" node to speed up the generation process.
-Slightly modified setting for better results.
FAQ
Comments (27)
Manager isn't finding ApplyFBCacheOnModel. Is there a step I'm missing?
it's in the description, you now need the WaveSpeed. There is a link in the description and follow their install instruction
@bissonfrederic69429聽 Knew I missed something. Thanks!
super neat workflow ! thanks for sharing !
Thank you. What is the best way to disable the upscale? When I bypass the nodes I get errors
bypass entire group
thank you for adding "Apply First Block Cache" to the workflow! really saves up a lot of time waiting for images ;)
i didn麓t see it before.
How can you take such a high quality screenshot of your full workflow? Do you have a 8K screen?
it's built in comfyui. right click and on bottom, export to png or something like that
Thanks for putting 'simple' in the title, just to make me feel stupid. No idea what this means:
NameError 89 WidgetToString
WidgetToString
Widget not found: 567.Number
Selects a node and it's specified widget and outputs the value as a string.
If no node id or title is provided it will use the 'any_input' link and use that node.
To see node id's, enable node id display from Manager badge menu.
Alternatively you can search with the node title. Node titles ONLY exist if they
are manually edited!
The 'any_input' is required for making sure the node you want the value from exists in the workflow.
edit: okay so works w/return all true, which would guess is not correct, but not sure what it is yet, ComfyUI Image Saver?
Is there a question in all of that? If you are starting using comfyui, you can simply use the workflow with lora that they provide in installation folder.
A question in all of that? I simply listed the errors in hopes of an explanation and help. Thus the info. First time Ive asked for help, but if you'd rather not
I've used Comfyui daily for about 2 months, along with Krita, so I am still learning. I've experimented with over 150 workflows, if that's relevant. Downloaded and run about 20TB in models and have 3 installations of Comfyui currently. I have run Flux, SD1.5, SDXL, and a few odd models, but mostly Pony and Illustrious. I do have a lot to learn, but I was joking about feeling stupid. To clarify, I got a WidgetToString error. I did not understand the explanation in the workflow for 'WidgetToString' or the error. I thought, possibly it was referring to the image saver. So, I guess my question was as clear to you as the error and notation on WidgetToString was to me. Moving on,, thanks for the workflow
@pumpher聽widgettostring is a node i use to convert any widget from a node to a string that can be used in the saved image metadata. Check the node being red after you hit the queue, it will be connected (the input) to another node. The widget of that other node must match the field in the node widgettostring. I remember a node that its widget was named "float" and after an update, it changed to "number". That was trigging a similar error to what you describe.
@bissonfrederic69429聽thanks problem solved
widgettostring error, solution for now is to name it Value
@pumpher聽I am having the same problem, how did you solve it ?
@amrpp79697聽in some comfy config, the widget (you know what is a widget right?) doesn't have the same name. Sometime it's "Number" someother time "Float". You must change the node widgettostring to reflect the widget of the connected node that you want.
@bissonfrederic69429聽Thanks
Hi, I'm having issue installing ComfyUI Easy Use for the easy string node (the import fails). What version do you use?
Use version 1.2s
@bissonfrederic69429聽Any one in particular? I've tried a few, from 1.2.0 all the way to 1.2.7, and I stilla have the issue, and 1.2.8 won't install properly.
the easy string node if supplied by ComfyUI-Easy-Use right? Any particular step to install the 1.2.8?
@tiftiffr761聽i mean, my workflow version 1.2s
Finally a workflow that actually works well.馃憫馃憫馃憫
With an RTX 3080 10GB, I run Flux.1 D + t5xxl_fp16 + clip_l + ae + 4 lora + upscale (1.4) to get a 1256 X 1616 photo.
#Prompt executed in 162.37 seconds#
WaveSpeed works really well decreasing the waiting times.
Congrats on your work and thanks for sharing.馃憣馃憣馃憣
Thanks
@bissonfrederic69429聽You're welcome!
I am nearly able to use this workflow! I am still missing the Image Saver node. In the workflow itself, this node appears as a small red box. When I select it, it expands to show what the node would look like if it is installed. However, it doesn't work, and I can't find a node that looks like this. Can you direct me, please?