I love the style of your videos, where you build the whole workflow again with us explaining what each node does and how it works this is significantly useful to understand the workflow so we can later spot errors and be able to fix them by ourselves if we ran into any error great work, I'm always waiting for your next video
I appreciate your comment; thanks a lot! I try to demystify ComfyUI for other people as its often a bit intimidating for newcomers. I'm still learning, so if you find any errors, please report back! I'm happy to learn and will make sure to correct the video if the error is to blatant.
Really glad to have style aligned in Comfy but Im running into a weird issue, when I generate 4 images(4 batch size), I basically get 2 groups. By that I mean that image 1+2 are the same, but 3+4 are a different style but the same as each other? So when I run 8 images(8 prompts, 8 batch), I still get two groups. 1-4 the same 5-8 the same? Really weird because everything is working and no errors appear in the command prompt. Anyone see something similar?
Good video. Thanks. WF links are not working, and not only under this video. I don't really understand the idea behind unsample/samplercustom. I don't believe it's doing some smart deconstruction of the original image, and it's just a regular img2img thing. Am I worng?
@@shubenoprod Yeah, I've actually fixed it and shared the workflow in the discord above. I'm not sure I'll get back at making Comfy Videos. I'm getting quite deep into Gaussian splitting, Unreal Engine and Spline. You guys interested? However, drop by the discord.
Does anyone else have problems loading the workflow into ComfyUI? I keep getting this error - When loading the graph, the following node types were not found: GlobalSeed //Inspire
Good video! I followed your updated workflow but keep getting this error: "Error occurred when executing BatchPromptScheduleEncodeSDXL: name 'pre_text_L' is not defined" Any ideas on how to get around this? Thank you
Thanks for the video and the flow! 1 question: why is the link you provided us a forked version of the Style Aligned ComfyU? It looks like the other one updatet the __init__.py file 2 weeks ago and in youre version it was last month.
Thanks, it worked great. One thing I noticed, often the generated images were not very good quality for example i generated a music instrument "sitar" which was completely broken in between. Also a character's leg was missing. How do we fix such issues after generation? I see, you are able to get very good quality images. Is it the model or something else because I almost matched everything like yours.
Now the "Style Aligned Reference Sampler" node takes in a "ref_latents" instead of "ref_latent" making it so you cannot connect the unsampler latent... any suggestions how to fix?
Maybe something went wrong when installing the node. It should accept "ref_latent" as you can see from my video. Perhaps try reinstalling the style transfer node.
When I try to run youre flow comfy wants the ref_latens(Referenz sampler). did the flow work for you without the connection? and thanks for youre Video!@@tech_bytebrain
I love the style of your videos, where you build the whole workflow again with us explaining what each node does and how it works
this is significantly useful to understand the workflow so we can later spot errors and be able to fix them by ourselves if we ran into any error
great work, I'm always waiting for your next video
I appreciate your comment; thanks a lot! I try to demystify ComfyUI for other people as its often a bit intimidating for newcomers. I'm still learning, so if you find any errors, please report back! I'm happy to learn and will make sure to correct the video if the error is to blatant.
3 minutes in and already love your video style...thanks for an excellent tutorial :)
This is so valuable. Just tried it and it is perfectly consistent. Thanks a lot!
Indeed ! It can also be combined with controlnet etc if you have anything specific in mind
Really glad to have style aligned in Comfy but Im running into a weird issue,
when I generate 4 images(4 batch size), I basically get 2 groups. By that I mean that image 1+2 are the same, but 3+4 are a different style but the same as each other?
So when I run 8 images(8 prompts, 8 batch), I still get two groups. 1-4 the same 5-8 the same?
Really weird because everything is working and no errors appear in the command prompt.
Anyone see something similar?
when i download workflow, "ref latents" in the style aligned box isnt connected to anything. there is no unsampler box and all my images come out gray
Good video. Thanks. WF links are not working, and not only under this video. I don't really understand the idea behind unsample/samplercustom. I don't believe it's doing some smart deconstruction of the original image, and it's just a regular img2img thing. Am I worng?
Thanks for the tutorial. I'm struggling with the ref_latent and the ref_latents on the StyleAlignedReferenceSampler panel.
Thats odd. You're not the first one to tell me that. Can you send me a screenshot on discord please? discord.gg/XSXMjGCm
hi! i got same issue, any solution fund here?
@@shubenoprod Yeah, I've actually fixed it and shared the workflow in the discord above. I'm not sure I'll get back at making Comfy Videos. I'm getting quite deep into Gaussian splitting, Unreal Engine and Spline. You guys interested? However, drop by the discord.
Does anyone else have problems loading the workflow into ComfyUI? I keep getting this error - When loading the graph, the following node types were not found: GlobalSeed //Inspire
Good video!
I followed your updated workflow but keep getting this error:
"Error occurred when executing BatchPromptScheduleEncodeSDXL:
name 'pre_text_L' is not defined"
Any ideas on how to get around this? Thank you
Yeah check out this updated workflow on my discord : discord.com/channels/1199741220235251884/1199741222324011160/1204463056491380736
@@tech_bytebrain thank you. Will check it out now
Thanks for the video and the flow! 1 question: why is the link you provided us a forked version of the Style Aligned ComfyU? It looks like the other one updatet the __init__.py file 2 weeks ago and in youre version it was last month.
Can you link the repo ? The one im using is the comfyui adaptation of style aligned. Feel free to refer me to a newer version. Happy to try it out
Thanks, it worked great. One thing I noticed, often the generated images were not very good quality for example i generated a music instrument "sitar" which was completely broken in between. Also a character's leg was missing. How do we fix such issues after generation? I see, you are able to get very good quality images. Is it the model or something else because I almost matched everything like yours.
Hey , If you've matched everything in my workflow its most likely the model itself. Are you using the SDXL base model?
Now the "Style Aligned Reference Sampler" node takes in a "ref_latents" instead of "ref_latent" making it so you cannot connect the unsampler latent... any suggestions how to fix?
Maybe something went wrong when installing the node. It should accept "ref_latent" as you can see from my video. Perhaps try reinstalling the style transfer node.
I'm using your workflow, if you Refresh the node do you see the "ref_latents"?@@tech_bytebrain
Currently away from pc ill get back at you tomorrow ;)
It works on my end , ive updated the workflow and replaced the unsampler. Try loading that one
When I try to run youre flow comfy wants the ref_latens(Referenz sampler). did the flow work for you without the connection? and thanks for youre Video!@@tech_bytebrain
thanks for the video. i'm wondering why you flipped the sigmas but you didn't connect the output to anything ...
They are connected to the scheduler right ? Am I doing it wrong ? 😑😂
@@tech_bytebrain The output of "FlipSigmas" node is not connected to anything, only the input of that node is connected, does it work like that?
@@b4ngo540 , @abdelrahmanabdolaid you guys are absolutely right! Simply delete the flipsigma node
So many error when executing Ksampler connected to style align
Such as ?