(Anima-1.1base/1.0base)✨基于anima-base-v1.0训练
❗请放弃使用1.0base模型，这其实是我在探索anima正式版模型的训练过程中所产生的，与1.1base版相比，1.0base算得上是个失败的模型。
请使用1.1base版本。以下说明均基于1.1base版。
增加了对furry的适配度
采样器我推荐er_sde/heun，调度器normal/beta/simple都可以尝试
如果出现对自然语言提示词不遵从的情况，可以适当拉高一点cfg值
当提示词比较多或线条轮廓不够清晰时，可以合理运用模型放大等技术
(Anima-v0.1)
🌟基于anima-preview3-base模型微调训练
关于anima模型的相关内容详见官方主页：Anima [官方] - 预览3-基础 |阿尼玛检查站 |Civitai
大概总结： unet模型而非checkpoint、对自然语言有更强的处理能力、与xl模型不同的采样器调度器设置（详见官方主页说明）
工作流可参考官方或将画廊中例图拖入comfyui即可
我做的事情一如既往。基于anima的初版模型，希望收到大家的反馈和建议！
Love you all❤❤！
(v2.0)
不同的美学呈现.
💡优化了1.5版本的角色大面积汗液问题；在小物体和细节上如花、蝴蝶等会有更好的表现；同样的，当像素量不够而出现糊脸等问题时，请利用好模型放大/高清修复技术。
(flashv1)
More dynamic ,More detailed, More landscape-optimized.
Illustration version，插画版
！需要注意，低分辨率下容易出现崩坏，请利用好放大工具（如ultimateSDUPscale）。经过我的测试，调度器选择normal、simple、beta都是不错的选择。例图均为beta所出。
(v1.5)
The new version is ready.
The new version has strengthened the male body and face, fine tuned the art style, and made the background cleaner, but some corresponding details may not be as rich as version 1.0.
新版本对男性身体和脸做了强化，微调了画风，背景更干净，但相应的一些细节可能没有1.0版本丰富。
Recommendations:
Sampler: euler a.
steps: 25-45 .
CFG Scale: 2-6.
Description
FAQ
Comments (20)
There're no mentions of e621 tags in the description, so I assume the Anima based versions of this model aren't trained with tags used in e621 [.] net.
yep. only selected the dataset and did some tuning.
求问如何可以压住皮肤上诡异的炫光，有的时候会很严重导致一眼ai
可以尝试提示词写的更详细些，或者直接调用画师风格/lora（不过我看了一圈目前没有发现相关lora）效果来得更快。如果有需要的话我过会可以训练一个此类型的Lora来改善该问题。❤
个人建议可以用厚涂相关lora进行减弱
@pesy dalao tql
May I ask approximately how many images were used for training?
Just about 300+
Ignores a lot of negative prompts, to the point that ComfyUI throws an error because it says the Clip Negative text is empty.
Are you sure it's not from your side? Just asking, cause I really want to get this model, but I don't want get errors.
@Emenimuno You can still try this model out, and test it out on a local Workflow like ComfyUI.
@bara_reader what I meant was, you're sure the error wasnt from some node that gave the error, and not the model? Cause I will try it, just figuring out why the error in the meanwhile.
@Emenimuno I am sure, since I tried multiple other checkpoints, and my negative prompt node worked. This checkpoint doesn't seem to be "all in one" complete, and you might need to supplement a negative prompt plugin with it.
@bara_reader it's beyond me what a negative prompt plugin is or how this is not a complete checkpoint, since I always get the base ones, without the AIO, but thank you for responding, got enough info to understand.
Are you using efficiency nodes? Anima is separate from unet and clip exactly.
@pesy I sadly was relying on the Load (all-in-one) Checkpoint. I now realized that these needs to be set of files and not just one. Can you add on the Description section what's your recommended clip and unet file?
@bara_reader so, it wasn't because of the model then, from what I'm reading. That's good. Cause I've tested it, and didn't run in any problems.