Same x.x

@Qvoheu It happens with versions 7 through 12 as well. Almost monochrome or very little color in outputs. Tested in SwarmUI.

@Qvoheu i posted some pics i was experimenting with, you can see that when i was removing the + version the color started to return, but sometimes i notice it happening with other embeddings too, and if it's not monochrome, it feels like result is being desaturated very hard

@Qvoheu also added a few recent ones i did just now, probably quality ones are doing this, and maybe it's reacting to the word "dark" that's the only guess i have

The latest updates to SwarmUI seem to have resolved the issue for me.

I use WAI V15, and occasionally I also get greyscale images with a black-and-white manga-like style. Using ‘greyscale’ as a negative prompt can prevent this.

Would be interested too. So I can remove unnecessary doubles.

depends on the embedding, there are some that include them (most "quality" embeddings) and others who dont (for some specific style, scenario etc).

@Qvoheu I am using it in the local ComfyUI and want to rename it to Chinese or familiar English. Otherwise, with too many files, I can't figure out the purpose of this file. Since it keeps fluctuating between good and bad, I'm not sure if it's related to my renaming of the file. Anyway, the logic of various software is more friendly to English. From the description in the post, it seems that I don't need to add the prefix "embedding" before the TAG.

If you're going to be sharing the prompts it might cause some confusion if you share the image +prompts to others without warning them, but for pure local use you can do whatever you want as long as you don't confuse yourself.

@LYNX233 This is a while after this comment but changing the name will actually be necessary for some embeddings sometimes, I've had the embeddings not work in comfyui because it was two words if you combine into one it will fix if that was ever and issue for you

with what model?

@alucardnoir941 it popped up with a couple of them. Smooth Noob and Illustrij. Though after I posted this comment they started popping up again. Been lowering the strengths of some Loras, which might be the real answer. It's been really frustrating to track down since it'll go away long enough to make me think I got it, then reappear later on.

@Qvoheu I've never used that word in either prompt, positive or negative. I am still beginning to think it was the strength of two Lora I was using. Knocked them both back, and haven't seen it pop up since, but we'll see. It's happened before where it went away and came back.

The best way in my experience to structure your prompts should generally be LoRAs first, then characters/descriptions, Scene, Art Style, Plain text describing the image, then Quality. So for example, it would be like
<LoRA>, <LoRA>, <LoRA>,
Solo, Saul Goodman, male, (suit, fixing tie), stern face, exhausted,
Office, [Court], Day,
Photo, (Realistic), Realistic Photo,
Saul Goodman standing in his office at the local Court, looking very stern as he adjusts his tie. His suit is slightly ruffled, and he's looking a bit exhausted.
High quality, Detailed, HighRes,

Generally speaking Stable Diffusion doesn't care how your text is actually formatted, but if you structure it that way, or similarly, it gives the best results. Thankfully, Stable diffusion also doesn't care if you mix plain text with your tags, and it can actually help describe the image you want to it. Just remember, Parentheses raise the emphasis on a portion, while square brackets lower the emphasis. More parentheses or brackets means even more or less emphasis. Stable Diffusion also loves descriptors like "Excessive" or "Tiny", so add descriptors where you can. You can also put them around groups of tags or text and it will emphasize everything within them.

@ThatBastardJohnny 我一般吧质量词放在lora后面，我听说越靠前的提示词权重越大，所以我觉得质量优先

Most of these "developers" just spam LoRAs, Checkpoints, Embeddings, etc to boost their rank on the website. Ego gratification and nothing else.

@LuminalArt they seem to be rewarded for it quite well.

Illustrious XL v2.0 are released under the CreativeML Open RAIL-M license. I would appreciate the poster to justify a 'no commercial' claim