v13e is basically the same as 13d but with IMG2IMG added and Detailer 3 is a different type of Detailer node.

I have most of the stuff bypassed by default since some people may not be comfortable with all the extra settings.

IDK about your default settings though?

@saytolol12 I don't really have a great way to explain it, but here's an explanation from Gemini:

Concat Conditioning (short for Concatenate Conditioning) in ComfyUI is a way to combine multiple prompt embeddings so the AI model treats them more as separate, distinct instructions rather than one blended, long prompt.

Think of it like this:

A Single Prompt (No Concat): If you tell a chef, "Make a red cake with blue frosting and a yellow sign," the chef might mix all those colors together into a messy, purple-ish blob. The concepts bleed into each other. The AI model's internal representation (the conditioning vector) often blends concepts when they are in a single prompt.

Concat Conditioning: You use the Conditioning (Concat) node to take the conditioning from two or more separate CLIP Text Encode nodes (or other conditioning sources) and literally stitch them together side-by-side.

You give the chef three separate, clear notes: "Note 1: Red cake." "Note 2: Blue frosting." "Note 3: Yellow sign."

The chef now has three distinct instructions to follow.

Key Benefits of Concat Conditioning

Prevents "Prompt Bleed": This is the most common reason people use it. It helps keep concepts separate. For example, if you have one prompt for a "red dress" and another for a "black dog," using Concat Conditioning helps prevent the dog from becoming slightly reddish or the dress from having black spots.

Bypasses the Token Limit (in a way): The standard text encoder has a limit (usually 75 or 77 tokens). Concatenating the embeddings from multiple text-encode nodes allows you to effectively provide more information to the model at the same step without averaging the results, as the model sees two distinct, full-length conditioning inputs.

Better for Multi-Subject Scenes: It’s especially useful for defining separate elements, like multiple characters, distinct foreground/background elements, or applying different concepts to different parts of the image with less blending.

In essence, you're telling the model, "Here is a primary instruction set (conditioning_to) AND here is an additional, separate instruction set (conditioning_from). Please consider both of these distinct entities."

@saytolol12 I haven't used region nodes, but from what I understand this is basically allowing us to add more prompts to give the actual prompt some "flavor". I just use it for lighting and stuff like that, but Gemini gave a more detailed description of how it works.

https://www.runcomfy.com/comfyui-nodes/ComfyUI/cfg-zero-star

Glad that worked for you. I am not sure why RES4LYF does that, but it has pretty much always been the common factor when folks run into that issue.

yeah, try turning on "Force Inpaint". That should force the detailer to do it.

Hard to say. If the background is bad before it goes into upscaling, then it's not likely to become amazing after. If it's decent before upscale and comes out bad, then you can try lowering the denoise, change the tile_padding to a bigger number, use a different sampler/scheduler, try a different upscale model, etc.

The downside of having a lot of options is that they all can affect how your image looks.

Try disabling Dynamic Thresholding, Detail Daemon, Epsilon Scaling, CFG Zero Star, and just run the workflow without them. If the backgrounds are still jacked, then it comes down to your sampler, scheduler, settings, and/or upscale model.

Some models also just suck at backgrounds.

Cool. Yeah, I encourage people to change/modify it to work what's best for them. There is no one-size-fits-all solution IMO.