Based on anima-base-v1.0, with fine-tuning and blending with a model by Fallen Incursio: Animice & Doe (Anima 1.0)
https://huggingface.co/FallenIncursio/Animice_and_Doe
The blending mode was DARE+TIES, using a personally created tool for Fine-Tuning and Merging models/LoRAs.
No additional styles have been merged into the model. Its goal is to remain flexible to various style LoRAs and authors, with improved prompt comprehension, detail, and handling through MWB enhancement.
The model is provided under the WhyxLicense
CircleStone Labs — complying with all permissions and restrictions of the original base model creator.
https://huggingface.co/circlestone-labs/Anima
___________________________________________________________________________________
ComfyUI - if you want to use it:
You can use my simple workflow for now, or any other:
AnimaWhyx - Fixed Upscale Workflow_v2
🎬 ComfyUI - Settings:
📸 CLIP SKIP: 2
🎲 CFG Scale: 4.5 | 5 | 5.5 | 6 (Optimal) / Shift 3.0-4.0 weight
🧼 Samplers / Schedulers: you can choose from er_sde (neutral and sharp), euler_a (softer, slightly more 2.5D), or dpmpp_2m_sde_gpu (more creative but wilder).
Hightly recommended: ER SDE
👣 Steps: 30 — 40
♨️ Hi-Res Fix: 4x-UltraSharp or Foolhardy Remacri (You can not use them, but with them more detailed images)
🎭 Adetailer: Not supported yet, use Upscale Nods.
📐 Recommended Resolution: 896x1152 - 1152x896 or 832x1216 - 1216x832, Simple: 1024x1024
✨ PROMPT PARAMETERS
✔️ Quality (Anima): masterpiece, best quality or masterpiece, best quality, score_7
(very aesthetic — optional)
___________________________________________________________________________________
UPD: 2026-6-29
Now Seaart.ai Support Generation - but you need to use Euler sampler - not Euler A!
Tested other Samplers what good work, but maybe give some noise. It's only for Seaart.ai on Anima with Generation Tab.
Samplers: Euler, DPM++ SDE, DPM++ 2S a, UniPC, PLMS, DPM adaptive
Euler - best of all.
Euler A - too smooth.
DPM series - small noise.
UniPC - Good.
PLMS - small noise.
Other very noisy and killed quality.
Description
🌟✨ What's New in v3 · v3.1 · v3.2 ✨🌟
🔄 Refined Blending Approach
The style transition is now softer and more delicate.
In previous versions, the merging was more aggressive, which sometimes caused certain styles to have a weaker influence.
Now the models are brought closer to a cleaner base, and they adapt to LoRAs significantly better.
╔══════════════════════════════════╗
🧪 Breakdown by version:
╚══════════════════════════════════╝
🎯 v3 — Minimal changes
• Close to *anima base v1.0*
• Enhanced details
• Maximum LoRA compatibility
🔥 v3.1 — Stronger emphasis on detailing
• More pronounced stylistic influence
• Better prompt comprehension
• Good LoRA compatibility
🎨 v3.2 — A focus on details and anime aesthetics
• Finest detail refinement
• Slightly more anime-oriented
• Improved prompt understanding
• Still great LoRA compatibility
╔══════════════════════════════════╗
🎭 Flexibility for any task:
╚══════════════════════════════════╝
Whether you want to see an author’s vision exactly as intended,
or plug in your own LoRAs — or even an entire collection —
these models are built to deliver.
The difference from v1 and v2 is truly substantial.
──────────── ✧ ────────────