Style Aligned ComfyUI Workflow: Consistency Across Generations ✨

In this tutorial, I'll walk you through a step-by-step process, using Style Aligned with SDXL in ComfyUI.

**Note : I forgot to lower the cfg scale on the Style-Aligned reference sampler for optimal results - USE UPDATED WORKFLOW / SamplerCustom Instead of Unsampler. **

Nodes used in this workflow:

Don't forget to subscribe for more valuable ComfyUI insights and tutorials. Let's embark on this educational journey together! 📚🎨

[bngo]

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

[samwalker]

3 minutes in and already love your video style...thanks for an excellent tutorial :)

[lmbits]

This is so valuable. Just tried it and it is perfectly consistent. Thanks a lot!

[xandervera]

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?

[pressrender_]

Thanks for the tutorial. I'm struggling with the ref_latent and the ref_latents on the StyleAlignedReferenceSampler panel.

[Rachelcenter]

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

[MrDhartz]

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

[TheNjordy]

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?

[abdelrahmanabdolaid]

thanks for the video. i'm wondering why you flipped the sigmas but you didn't connect the output to anything ...

[PallaviChauhan]

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.

[SejalDatta-lu]

Good video!

I followed your updated workflow but keep getting this error:
"Error occurred when executing

name 'pre_text_L' is not defined"

Any ideas on how to get around this? Thank you

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?

[StableFun]

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.

[RahulGupta]

So many error when executing Ksampler connected to style align