NoobAI-XL (NAI-XL) Reviews

Usually I wait and test models thoroughly to leave a review. But since this model has sparked some controversy: I have decided to review earlier than usual. Firstly: If you know your way around Danbooru and e621 tags: this model works as advertised. It is noob friendly if you know tags well. It doesn't do everything, but it captures a vast amount of concepts. This model is particularly excellent with clothes and understands most tags well. Despite the stark complaints this model has received: I have tested it thoroughly and found it to be an excellent model. People complain about the high contrast. I prefer high contrast on the images because I can control it with simple (high contrast:0.7) in the neg and lowering the CFG, where as in other models: no matter how much I try to increase the contrast: it always ends up looking dull and with low contrast and saturation. Some people complain about the japanese text. It just so happens that it is a tag and one can put "japanese text" in the neg and that works. The "solo" tag in the positive prompt removes multiple scenes and "comic" in the neg goes a long way. This version reduces the quality of furry-style artists severely (compared to version 0.75), but it does perform well with Kemono-style artists (japanese furry) which is not something that other furry models do well. It is not an end-all-be-all model, but it does what it does excellently. I understand how much work and effort it takes to create a model such as this. Truth be told: this model is a must have in every AI artist's toolbox. One can also have version 0.75 alongside this model and use one for some things and another for others. An artist that knows what s/he is doing does not use the same brush for every stroke. Likewise: one must change AI models depending on the image and composition one wishes to achieve. One final note: for the people wishing to do NSFW art: "uncensored" in the positive prompt goes a long way, along with "explicit". Also use "questionable" (for those of us wishing to generate with clothes doing some appealing eye candy) and "safe" (for desirable images which we can share with anyone without remorse). These rating tags truly help composition if used wisely and can single-handedly add much quality to the image. "masterpiece", "good quality", etc. Also add some extra spice to the images. Make sure to add them as appropriate and experiment with using lower qualities (they also improve composition depending on what one is doing). "worst quality" can add some charm when used in the positive prompt (especially if you know what you are doing). To the noob team: keep doing what you are doing. You are doing excellently. This is a must have anime model which I have been waiting for since waifu was released back in the SD1.5 days. If at all possible: please appease the AI community by publising the intermediate training models like you did for the first half of the project. I am looking forward eagerly to test them out. Many Blessings to all.