模型更新内容 | Update notes:
新版本模型 Fn‑Base‑Anima‑V1.0，针对大量用户想要非 turbo 版本的需求开发。样图虽用了 turbo LoRA 生成，但实测该模型脱离 turbo LoRA 依旧可以产出高质量图像.
新版本模型 Fn‑Base‑Anima‑V1.0 侧重多画师画风还原。样例图单独使用
@deluxe<<<,@imamura ryou,@ogre \(illustogre\)等画师标签。原生 Anima 对多画师标签适配不佳，本模型着重提取固化画师画风特征。个人实测整体表现优秀；部分画师画风输出偏弱时，可上调标签权重。推荐使用turbo LoRA获得与例图相同的效果.
如果你喜欢我的模型，不妨点个赞支持一下；如果愿意分享你创作的作品，或是提出对模型的建议，那就再好不过了。感谢大家！
The new model, Fn‑Base‑Anima‑V1.0, is developed in response to numerous user requests for a non‑turbo version. Although the sample images were generated using the turbo LoRA, practical tests show this model can still produce high‑quality images without the turbo LoRA.
The new Fn‑Base‑Anima‑V1.0 focuses on replicating the art styles of multiple illustrators. Sample images use standalone artist tags such as @deluxe<<<, @imamura ryou, and @ogre (illustogre). The original Anima model has poor compatibility with multi‑illustrator tags; this model prioritizes extracting and embedding stylistic features of various artists. Personal testing demonstrates strong overall performance. If certain illustrator styles render weakly, you may increase the weight of the corresponding tags.
It is recommended to use turbo lora to achieve the same effect as the example image.
If you like my model, please consider leaving a like to support me. It would be even better if you’re willing to share the images you create or offer suggestions for the model. Thank you all!
模型介绍 | Introduction:
本模型基于 Anima BASE1.0 模型, 融合多个个人训练LoRa以及Tubro LoRa调整生成.
本模型不需要负面提示词.
当前所有测试皆基于Comfyui, 未在Webui中测试.
受训练素材影响, 模型生成角色可能在无 'halo' 提示词情况下出现光环, 可在提示词中添加 'no halo, no hat' 以解决.
This model is based on the Anima BASE1.0 model, generated by integrating multiple personally trained LoRAs for adjustment.
This model does not require negative prompts.
All current tests are conducted based on Comfyui and have not been tested in Webui.
Affected by training materials, the model may generate characters with halos even without the 'halo' prompt, which can be resolved by adding 'no halo, no hat' to the prompt.
参数推荐 | Recommend：
Resolution: 832*1216/1024*1536
Steps: 12-16
CFG: 1/1.2-1.4 (Adjust the CFG to 1.2–1.4 to enable negative prompts.)
Sampler: er_sde/euler/euler_ancestral
Scheduler: beta/simple
Hires Fix Step: 5
Hires Fix CFG: 1
Hires Fix Sample: er_sde/euler/euler_ancestral
Hires Fix Scheduler: beta/simple
Hires Fix Denoise: 0.5
Description
3.1 版本是目前为止我最满意的一版。它在保留原有动漫风格的基础上，大幅强化了背景层次与材质质感，整体画面依旧干净通透。尤其在人物皮肤、丝绸、丝袜等细节材质的表现上，效果尤为出色。
说实话，这个模型我其实很舍不得公开，但还是希望能让更多人体验到这份成果。如果你喜欢我的模型，不妨点个赞支持一下；如果愿意分享你创作的作品，或是提出对模型的建议，那就再好不过了。感谢大家！
Version 3.1 is by far my most satisfying release to date. While retaining the original anime style, it greatly enhances background depth and material textures, keeping the overall visuals clean and clear. The rendering of detailed materials—such as human skin, silk, and stockings—is particularly outstanding.
To be honest, I was quite reluctant to release this model to the public, but I still hope more people can experience this work. If you like my model, please consider leaving a like to support me. It would be even better if you’re willing to share the images you create or offer suggestions for the model. Thank you all!
FAQ
Comments (9)
The model looks beautiful, but if only you could make it without turbo... Otherwise, without negative prompts - using the model is masochism
Big fan of the skin and materials on this one!
[Look inside]
[See no prompts.]
[Don't download.]
If you go to my profile and check the Announcement, you will see I wrote: "If you need a prompt for a specific image, please leave a comment under the picture, and I will send you the prompt in reply."
@fannon Didn't see that. Thank you.
what's the prompt for mage girl preview image?
P.S. It would be great if you make a non-turbo ver. 'cause turbo is very strict about face expression and less flexible than base ver.
The prompt for the image you mentioned has already been added.
In fact, images generated by the Turbo model may have less variation, yet they are more stable in my experience. Most visual outcomes can be achieved by adjusting the prompts. To introduce randomness, I use tools like Z-Tipo on the base prompt to generate diverse results.
Additionally, non-Turbo models produce visuals with a noticeably different style from Turbo models; a perfect match cannot be achieved. The artistic style also tends to be inconsistent, largely affected by varying negative prompts. What’s more, each generation takes a longer time. I prefer generating high-quality artwork in a shorter timeframe. If you need to use negative prompts, try setting the CFG scale to 1.4 or enable NegPip.
As for what you mentioned regarding facial expressions, I’m not entirely sure what issue you’re referring to. Could you provide an example if possible? It might be a problem I haven’t noticed yet.