Note: I spend a lot of time on these to make sure they are top notch. If you have ANY issues, I'm very responsive. hit me up and I'll resolve them for you.
This is an advanced workflow. It is node heavy and designed for multiple options and to return professional high quality results. If you are looking for simpler workflows, they are here:
An essentials one can be found here: https://civarchive.com/models/2349309/illustriousponysdxl-essentials
A simple one can be found here:
A complete user and troubleshooting guide inclusing a list of common error messages can be found here https://civarchive.com/articles/24678/how-to-use-my-workflows-a-comprehensive-breakdown
Instagram: https://www.instagram.com/synth.studio.models/
Buy me a☕ https://ko-fi.com/lonecatone
This represents hundreds of hours of work that I give away for free. If you enjoy it, Please make sure to 👍like and post to my workflows. That is how you support me. Also, feel free to ⚡tip 😉
也有中文说明
VER 10.1:
Got tired of having to always mess with resolution, so I gave you the option to make it automatic
Math node replacements to deal with the Linux errors
Added psst... fast bypassers (you'll thank me).
💡use the upgraded face yolo I added. much more accurate.
Improved notes throughout
Ver 8.0/ 9.0:
Major overhaul to bring it up to speed with my other workflows
Added second checkpoint metadata
added a prompt to hi-res fix
added Inpainting suite from the pro grade workflow by request.
V7.0
Apparantly , I forgot to hook up Face Swap in my last model and nobody told me until today, so a free major upgrade it is!
Installed a sentence remover in the prompt assist area to help eliminate CivitAI 🚩
Cleaned up some bugs
Resinstalled Qwen VL Prompt Enhancer
Upgraded Face Swap
also upgraded notes on how to remove it.
Minor changes to detailers
Added an inline crop tool in the post production suite.
Ver 6.0:
Major overhaul to bring it up to speed with my other workflows
Added a refiner checkpoint that can easily be switched on or off
Added FaceID as well as a style and composition IP Adapter for character consistency
Reowked the Detailers ro improve them
better version of Hi Rez fix
Reworked the Save with Metadata to align with CivitAi and fix the Comfy update bug 🙄
This is an all in one workflow:
Controlnet
I2I with prompt generation
Dual Model Loader: Better control
Sage Attention: Speeds up production
Ultimate upscaler: For prescale and hi-res fix
Detailers: Face, eyes, hand, expression
NSFW Detailers: Male & Female anatomy
Seed VR2 Upscaler: Get up to 4k pics, even on low VRAM
Post production suite: My pride and joy. Does everything.
Description
Improved Controlnet
Added Qwen VL Prompt enhancer
Added a JPEG scrubber to the Post Processing Suite
Improved some math functions
FAQ
Comments (43)
I'm a beginner, and this is the most complex workflow I've ever seen, but I will try my best to overcome this and understand it.
@zlikk602596259999 Then this is designed for you! There si a very helpful instruction guide, and feel free to ask questions.
A lot of errors and I only see a part of the workflow:
TypeError: link_info.resolve is not a function
at SetNode.onConnectionsChange (http://127.0.0.1:8188/extensions/ComfyUI-KJNodes-main/js/setgetnodes.js:106:33)
at SetNode.configure (http://127.0.0.1:8188/assets/index-DIU5yZe9.js:35130:35)
at LGraph.configure (http://127.0.0.1:8188/assets/index-DIU5yZe9.js:44993:17)
at LGraph$1.configure (http://127.0.0.1:8188/assets/index-DIU5yZe9.js:120908:26)
at LGraph.configure (http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/reroutePrimitive.js:14:29)
at ComfyApp.loadGraphData (http://127.0.0.1:8188/assets/index-DIU5yZe9.js:121365:18)
at async app.loadGraphData (http://127.0.0.1:8188/extensions/ComfyUI-Manager_DISABLED/components-manager.js:774:9)
at async reader.onload (http://127.0.0.1:8188/assets/index-DIU5yZe9.js:121738:11)
are you on desktop? This is a first
@Moverx yeah, you need to update ComfyUi and the nodes. That is a getset issue.
@lonecatone23 And that's exactly what I'm afraid of. I've created a machine on a USB drive where everything works for what I'm doing, but all updates are blocked. After the update, half of the things might not work... eh
@Moverx you should be on portable. Not desktop. Nothing new is going to work for you
I'm pretty proud of myself i actually got it working haha. Total noob. what i did was put all the errors into google ai and it told me what to do. i pretty much nuked my install, downloaded pytoch, hsome visualizerc++, had to downgrad emy transformers from 5.0 and then i feed the errors into the ai and it told me how to fix em lol. Now to figure the whole vl thing and I think i can use things thing. :) i think it downloads the files or soemthing. Just posting this as i i think I upgraded my power level in getitng comfy ui to work hehe. :)
btw where do you put the lut files lol
@kevfactor859 There's a file location map in the notes. Good job! This is a good workflow to learn from.
@lonecatone23 Thanks! i changed the loader to gguf too. I had trouble with qwen 3 so I use 2.5. it was a a lot of dos stuff i didn't understand but gpt seemed to fix most of it hehe.
BTW just curious about some of the stuff with controlnet and mask. past getting a prompt im not sure what the use of those are yet. does it do image to image? I guess you turn off most the other stuff when you use em too :)
@kevfactor859 so controlnet and prompts can doa couple of things. You can load an image, turn on Florence, and t will create a prompt based on that image. Controlnet will duplicate teh pose esactly. You have several options, but DepthAything and Canny are the best
man i cant install that i2i node, i'm trying that Miaoshouai-Tagger, i copied the custom node, but comfyui still doesn't detect it. anyone knows how to make it work?
are you on Desktop? update your comfyUi main package
@lonecatone23 yeah i managed to install it and make it work (the tagger node), i had to install flash attn. but now i'm having other issues, it says i need these nodes,
SDAnyConverter
RemoveDuplicateTags|LP
AILab_QwenVL_PromptEnhancer
But as soon as i enter the manager to look for missing nodes i get this error,
Failed to find the following ComfyRegistry list. The cache may be outdated, or the nodes may have been removed from ComfyRegistry.
comfyui-qwenvl
The others say something like IMPORT FAILED , and some errors :S maybe i need to use an older version?
@raidou88 are you on Desktop or portable? If desktoip, you may wasnt to run the essentials workflow as they haven't updated ComfyUI
The best solution is to take your log and feed it thourgh Grok or Claude to resolve the issues
@lonecatone23 portable version. i managed to install some nodes, i still have some troubles with seedvr that i didn't try yet because i feel that will ruin another workflows i have, so ill bypass it for now..
Some issues i had and ill write in case someone has the same problems,
Failed to find ComfyRegistry list – comfyui-qwenvl > i needed to install llama-cpp-python, some kind of library i had to download. had to downgrad nunpy
other error was
ModuleNotFoundError: stable_diffusion_prompt_reader
solution:
git clone --recurse-submodules https://github.com/receyuki/comfyui-prompt-reader-node.git
then wordninja
Installed wordninja manually by creating wordninja.py in site-packages and using the wordninja_words.txt.gz dictionary with a compatible loader.
Now the problems I still Have, ailabqwen vl doesn't work, even if the modules are installed, it only works with the gguf version. and when i try to run the workflow using them i get this vae error.
^^^^^^^^^
RuntimeError: Given groups=1, weight of size [512, 16, 3, 3], expected input[1, 4, 168, 135] to have 16 channels, but got 4 channels instead
Prompt executed in 247.15 seconds
gotta try tomorrow :/
@raidou88 Okay, teh registry list is a very common error to comfy. When you update, it will use the old log, and it won't rewrite. Open teh terminal, make sure all startup tasks are complete, then start again.
you CANNOT mix quants. In other words, no gguf clip loader and reguklar diffusion model, or using Qwen_3_4b instead of Qwen_3_8b for Z-Image Base (or vise vesa foir ZIT). That is your weight error.
QwenVL requires an older version of pyTorch. Your computer is probably running of of Torch 5.0, it requires an earliler version. I'm going to strip the whole POS out of the next pass of wporkflow updates. I frigging hate it. Just del;ete it, and the "join strings" node and teh workflow will heal itself. Trust me, you're not missing a thing. I just liked it because it was one of the few LLM promp[ters that didn't require an API token
@lonecatone23 i had a nasty bug that only showed gguf versions of Qwenvl, it didn't show me the ones you had in your workflow but i fixed it. now i think (i hope) that i can learn how to use it haha.
For anyone on Comfy Portable on Windows, open a command window and go to the "python_embeded" directory then run the command:
.\python -m pip install transformers==4.57.0
that fixed it. i had transformers 5
@raidou88 the error and fix were already listed in my troubleshooting guide here: https://civitai.com/articles/24678/how-to-use-my-workflows-a-comprehensive-breakdown-and-troubleshooting-guide
Thanks
@lonecatone23 thanks, i can make some images but i don't know why it says the job failed. The only reason i can think of (because its on a red square) is SAMloader (impact), it said sam_vit_b_01ec64 (but was only the name it came with, it never loaded it).
I downloaded the SAM3 custom node, i got the access to the repository, i downloaded sam3.pt i put it in a sam3 folders that i created, inside models. i even looked for sam_vit_b_01ec64 on the web , and put it together, but still sam loader won't detect anything. now it says NULL. i know it says it's outdated, but the loader not detecting the files is it normal?
@raidou88 man, you're full of new issues 😋
If your sam model loads, then check which detection models you are using. I believe I listed my favorites somewhere
eyeful. pt
face_yolo8m
etc.
I don;t know how mauch modifications you've done to my workflow, but you may want to just reopen the original and compare
Also feel free to message me directly if you need to.
@lonecatone23 (Edit: i fixed it downgrading impact to an older version. now proceeding to try and make florence 2 work )
Oh btw, Florence2 was giving me problems saying that none wasn't a model or something like that. wasn't working. so i disabled that get_flo mod that was linked to florence2run, and added the florence 2 model loader, and now i can see the tags
@raidou88 the florence 2 run model downloads itself 🤔
@lonecatone23 well my comfyui must have been really messed up and i'm glueing it up back again. haha, florence2 didn't want to work until i did that.
i struggle with every module . now it's the turn of qwenvl prompt enhancer that does nothing, it just stays at 0/10 steps and doesn't do anything and i have to restart comfyui. I'm gonna test it in another workflow to see if i can make it work
@raidou88 turn everything off but save draft. You're the only person I've heard who has these issues like that. It's almost always a desktop issue
@lonecatone23 i left it like 30 minutes and nothing, i had to restart comfy, but now i downloaded the workflow from qwenvl github to test if that would work, and took like 5 minutes without showing anything, but then it worked. now it takes a few seconds and works on your workflow, not sure what fixed it because the qwenvl model was downloaded (so it was nothing new).
Thanks, I'm finally using it, i can make some good results and others not much. haha
I wanted to ask, i was using the i2i module, and the results were good, but the picture seems cropped and zoomed in , in comparison with the original. is that a ksampler thing or do i have to change somethign?
And 1 last thing, why the compare module wont compare the draft and the post processed (?)
@raidou88 the dimensions are set from the aspect ratio node. There shold be no issues wit hthe image compare node if all of the images show up in teh post processing suite. If they do not, then you aren;t processing beyond teh draft switch
@lonecatone23 i press the switch on the draft node the one that looks like a fast forward button, but i think it starts again from 0, it never reaches the post processing group. what do i have to activate?? (O_O;)
@raidou88 read the instructions. If you don't set the seed, it will start again, or just turn it off.
Read the instructon manual here. ALmost everything you have asked I have listed out.
https://civitai.com/articles/24678/how-to-use-my-workflows-a-comprehensive-breakdown-and-troubleshooting-guide
@lonecatone23 yeah i was reading the instructions, but it was kind of confusing, i kept pressing the button on draft mode that looked like a fast forward ►► when it should have been the middle one to just disable it.
But still i got (another error) that is not in your manual T_T (i got that error on female parts detailer)
DetailerForEach
'DifferentialDiffusion' object has no attribute 'apply'
so i disabled it, and now i can see the image loads on some detailers but the next one that fails is the face one. exactly like the female parts.
FaceDetailer
'DifferentialDiffusion' object has no attribute 'apply'
you wrote about oom on those detailers but this doesn't look like that.
i'm using faceyolov8m-seg.pt and femalebodydetection_yolo26 i tried others detailers files, but still have the same issue with those, should i keep looking for segm models?
@raidou88 this is an impact pack. update. I hae said several time you need to update.
if it still gives you issues, slap a differential diffusion node between it and the model
I have 1 simple question. Does this workflow do corn? no sfw or n5fw/borderline corn. Straight up corn?
That is not a factor of a workflow, but a checkpoint, type of model, Loras, and a prompt.
Yes, this does whatever you enter into it
@lonecatone23 your work is amazing, i have a question bro im now in the part of nsfw creation im struggling to make character with consistent pussy and asshle can you recommend some checkpoint, model, etc for that? btw im using zit. thanks in advance bro!
@miracc0014976 Yeah, don't use ZIT. Zit sucks for that.
Use Loras like Mystic ZIT, Turbopussy, better females, etc. Thern use teh detailkers to clean it up. Be aware that you'll probably genetrate 4-5 pics for each usable one, so running a random seed with draft mode off is recommended
- what do you like about corn?
+ It's corn!
Sorry, couldn't help myself.
don't know about corn, but it peas sometimes 😊
@emotionaldreams2 that's not permitted here, nor are bean byproducts
@lonecatone23 ..it was a joke. a play on corn 👍
@emotionaldreams2 Lol, I know "Peeing" is banned on this site