✨ BAS – Better Anime Style Model for Anima ✨
BAS (Better Anime Style Model) is a merged checkpoint based on Anima Base v1.0, designed to enhance and reinforce a modern anime-style aesthetic.
This checkpoint responds particularly well to prompts such as:
anime, anime_coloring, anime_screenshotRecommended Settings
Lora: Anima Turbo LoRA
CFG Scale: 1
Steps: 8
Sampler: Euler
Using this model together with "Anima Turbo LoRA" may produce better results.
X ( Follow me on X for updates on what I'm working on!🙇 )
https://x.com/news_v_com
License
This model follows the same license and usage terms as Anima Base v1.0. Please refer to the original Anima Base v1.0 license for details.
Description
FAQ
Comments (19)
Just found my favorite Anima checkpoint.
Thanks so much!🙇
I’m really impressed by how well this checkpoint synergizes with Anima Turbo LoRA. I genuinely think it’s one of the best 2D anime models available right now.
Thanks so much!!🙇
After some frustating round of test and tweaking, this is my finding about this checkpoint:
1. This model requires the Anima turbo lora for the result to have at least some semblance to the samples. Without it, the result might have more details which in effect makes it looks less anime. This kinda make sense since anime aesthetic usually rather plain in detail and Turbo mode less application on detail might have helped to achieve this anime looks better. Which also the reason why detailer don’t work well with this model since the added detail diminish its anime looks.
2. The reliance on the turbo Lora seems greatly diminish this checkpoint prompt adherence, especially when it involved multiple characters. This model only creates junks from my threesome scene prompt which other model have no problem with. Threesome scene with simpler prompt sometimes work, but yeah sometimes while other has no problem at all.
3. One might wonder then why not merged the Turbo lora directly into the checkpoint? While the Turbo lora is very integral to this model base style and capability, not merging the Turbo lora provides versatility. Even without the anime related tags this model base style is nice and unique enough that you might want to use it in conjunction with style loras.
4, The base style of this model—when used without the anime related tag—has more mature looking character, which unexpected but not surprising. The samples provided here shows characters that not only has looks in anime style, but also more mature. This is why not merging the Turbo lora directly provides option since this base style (for now) is unique to this model and people might to use it in different way.
My suggestion for future version would be: 1) keep the base style looks and its anime style capability since this is its strongest point 2) fine tune its performance and stability when not used in conjunction with Turbo lora 3) improve its prompt adherence especially natural language prompt.
I really the base style and looks of this model, so I really want it to improve.
Thank you so much for your kind, detailed comment and suggestions!🙇
To be honest, I had never even considered merging the Anima Turbo LoRA into the checkpoint. I'll definitely give it a try in the next version.
The same goes for your point about Detailer. I'll look into it and do some testing to see how it affects the model.
Your feedback was incredibly helpful because it highlighted things I would never have noticed on my own. It was mind-blowing.
I'll use everything you've shared as a reference while refining and improving this checkpoint.
Thank you again for taking the time to write such thoughtful feedback. It really helped me a lot!🙏
@news_v_com In regards to Turbo lora, you shouldn't merged it. Because like I said, the option is nice when you keep it separate from the checkpoint. Complication might arise if you merging it. My suggestion is that you should write it down in the description that you used Turbo lora to achieve the results as show by your samples since the effect is significant.
Also similarly with the detailer, it isn't that it doesn't work but rather using it will affect the style and the result might ended up different from the samples (aka not anime-like enough). You should just mention this in your description (though I wonder if it's possible to make detailer works with the anime aesthetic).
Personally rather than merging the Turbo, you should fine tune the model so it could achieve the anime-look without it. Since from my testing, not using Turbo lora not only makes the looks significantly less anime but also a lot more unstable.
@Thongor Thank you for your kind and thoughtful comment!🙇
You're absolutely right—I should have mentioned the Anima Turbo LoRA requirements in the description from the beginning. I'll update the description to make that clearer.
Since I'm still using an old RTX 3060, I haven't done much testing without the Anima Turbo LoRA. Your feedback made me realize that I need to spend more time testing and tuning the model so it can achieve an anime-style look even without Turbo.
I'll work on improving it and release an updated version soon!
Thanks again!🙏
這不錯!比其他來說更容易還原動漫原圖!
非常感謝您的留言！🙇
I don't really know how to use these models, can you tell me which clip and vae should I use?
Thank you for your comment!🙇
The usage instructions are explained on the Anima page on Civitai.
https://civitai.com/models/2458426/anima
Since this model is based on Anima Base v1.0, it can be used in the same way as Anima.
@news_v_com thanks for the reply! keep the good work fr
大佬，我怎么生成不了你封面的图片，我的模型和VAE及CLIP也按照你推荐的，但是生的图片完全不是一个东西，这是怎么回事，请指教
感谢你的评论！😊
请确认已经在 LoRA Loader（LoRA 节点）中正确加载了 Anima Turbo LoRA，并将 CFG 设置为 1。这样生成出来的图片应该会和示例图比较接近。
另外，我还推荐你试试另一个版本——“BAS+ Better Anime Style Plus”。使用这个版本的话，通常会更容易生成和示例图相似的插画。
如果方便的话，欢迎试试看！希望能帮到你。祝你创作顺利，好运！🍀✨
@news_v_com 谢谢大佬解惑，现在出图没有问题了，非常感谢，但是用加速LOEA的话质量就不太行，脸还可以看，手就糊成一坨了
內置TURBO LORA 我覺得可以接受...但顏色上有點2010年的風格偏淡...而且手指會糊化及很易6隻手指