"best quality" and " best quality" (with a leading space) produce different stylistic tendencies in the generated images.
A single quality tag like "best quality" vs " best quality" shows only a subtle difference, but when you compare
"year 2025,year 2024,newest,masterpiece,best quality,score_9,absurdres"
(no spaces)
with
"year 2025, year 2024, newest, masterpiece, best quality, score_9, absurdres"
(spaces after commas), the generated art styles are obviously quite different.
Prompts without leading spaces before tags generally produce flatter, more planar, and lower-quality-looking images.
Although the precise mechanism is unknown, the leading space likely isn't just random noise — it probably makes the model interpret " best quality" as a meaningfully different tag from "best quality".
It might just be a placebo effect, but give it a try yourself — many people seem to notice a clear shift in the overall art style tendencies.
I've noticed images like that on models like Illustrious where they don't have any spaces to be honest with that model, I don't think it makes too much of a difference, but with a model like this it probably is a little bit more sensitive. I'm still yet to find the perfect prompt. If you are looking for a bit more stability with images I would check out the checkpoints as well. My favourite one at the moment is AnimaYume As regarding anatomy and stuff like that, it's more stable and doesn't give you wild results as much. It's obviously still a preview until we get the full model. I think these stuff should be ironed out by then. Exciting times with this model. It's a very good model.
I've noticed images like that on models like Illustrious where they don't have any spaces to be honest with that model, I don't think it makes too much of a difference, but with a model like this it probably is a little bit more sensitive. I'm still yet to find the perfect prompt. If you are looking for a bit more stability with images I would check out the checkpoints as well. My favourite one at the moment is AnimaYume As regarding anatomy and stuff like that, it's more stable and doesn't give you wild results as much. It's obviously still a preview until we get the full model. I think these stuff should be ironed out by then. Exciting times with this model. It's a very good model.
I also like AnimaYume. And even then, the art style changes depending on whether there's a space after the comma. Without the space, the art quality does deteriorate.
With SDXL, spaces after commas don't matter at all. They tokenize to exactly the same thing. For LLMs, it does matter. You should use the format with spaces after commas.
It has been a long time since I tested it, but if you shove a huge tag list with no spaces at all, the quality should be a LOT worse. Something to keep in mind especially if you are copying prompts used on SDXL models. A lot of people don't use spaces there or use them inconsistently because it doesn't matter for SDXL.
Does this also apply for line breaks as well? e.g. when writing natural language prompts, maintaining one paragraph versus multiple short paragraphs.
Don't line break, seems better. Do test it yourself.
I created a formatting node for Anima so I can keep my prompting habit from SDXL: https://github.com/1lch2/ComfyUI-AnimaPromptFormatter