Anima official is still in preview — no stable release yet.
This is a Pre-alpha v0.1 release — expect rough edges and some instability.
Beta is switch to the anima beta version 1.0 of the base checkpoint.
Recommended settings
Resolution: txt2img at 720p or higher, with Hires fix targeting 1080p+
Hires upscaler: Remacri (recommended)
Steps: 20 - 50 (reference:35)
CFG Scale: 4–6
Sampler:
er_sde,euler_a, ordpmpp_2m_sde_gpuSchedule: SGM Uniform, Uniform, or Beta
Compatibility: Currently runs locally only on Forge Neo or ComfyUI.
VAE is baked into the checkpoint — no separate VAE file needed.
This checkpoint is in slow development — updates will roll out gradually, and may pick up pace following new releases of the official Anima base checkpoint.
Forge Neo settings reference:
Description
This is a Pre-alpha v0.1 release — expect rough edges and some instability.
FAQ
Comments (13)
I can't believe there's already a furry model for Anima so soon! Thank you so much for the quick release! ~
This model is good look at what i made with some Lora https://tensorhub.art/images/992100214720431055
I should probably start learning Anima — people are making better art with my checkpoint than I am. 😅
After all, a model still needs proper showcase images.
The advantages are very obvious. Even if you ignore its ability to write text, the improvement in image composition is nothing short of a qualitative leap. I can even have it generate a split pose in a vertical image—something that was completely impossible in the illustrious/noob era.
The biggest difference between NoobAI/illustrious and Anima is that they're from different architectural generations. NoobAI is built on SDXL, while Anima is based on NVIDIA's new Cosmos architecture, architecturally similar to Flux (its image generation logic can be thought of as a lightweight, anime-focused version of Flux). Anima has a newer architecture with higher potential, but it's still on its third preview build (preview 3). There's no ControlNet yet, and the ecosystem is tiny, with barely any LoRAs to work with. The final release could end up quite different from the preview versions, or might never make it out of preview at all.
What's new about this furry model
Can you make furry styles with anima 3 This will be better
感觉很棒的ε(´｡•᎑•`)っ♡
Which VAE is built-in, the fine-tuned one or the Qwen one? I can't find a method in Comfyui to call the built-in VAE of this model
看到 Anima base 1.0 底模发布了，就在想作者大大啥时候能更新一波新的 Furry Anima。初版使用了很久，效果确实很好。不过作为初代，用久了确实也遇到不少问题，想反馈一下（仅代表个人使用体验）：
1. 物种多样性存在不足：感觉训练集里小众物种少了点，比如鸟类大多数跑不出来。部分犬种也有点难，像伯恩山这种，哪怕指定了毛色也还是不对；金毛的脸看着有点“恐怖谷”，雪豹的脸也有点怪。
2. 过拟合：如果不特意描述外貌，只写 wolf boy 的话，生成的狼人面部毛色基本都“撞脸”了。
3. 一些细节 Bug：画老虎的时候，嘴角到下巴那一块经常莫名其妙出现黑色色块；另外感觉肢体比例出错的概率比 XL 模型要稍微高那么一点点。
再次感谢作者的辛苦制作！真的超级期待新版发布（绝对不是催更哈，但我做梦都想明天就能玩上新的Furry Anima，手动狗头）。
Hi all!
請問哪邊可以安裝Forge 的sampler:ER SDE呢?
where can I install the sampler er sde for Forge?
You might just need to update your install, mine is months old and I have it and I never had to go install that sampler or get it separately.
@Straitjacket thanks! I figure out that I was using forge classic, the ER SDE should be found in Forge Neo.