Anima 2B z3v0.1 is a style merge on the 2B Anima architecture. It is Dreamland ANM v0.1 with five Anima style LoRAs baked in at fixed weights, so the look is there without stacking LoRAs at run time. It aims at painterly anime illustration with atmospheric depth, sourced light, and clean line work.

It is a standard 28-block Anima diffusion model: it loads like any other Anima checkpoint and works with Anima LoRAs trained on the 2B base.

Recommended settings

• Steps: 30 (30 and 40 look the same; 30 to 50 all fine)

• CFG: 4.5 (4 to 6 all clean). Negative prompts work at this CFG.

• Sampler / scheduler: er_sde / simple

• Model sampling: AuraFlow shift 3

• Resolution: about 1 megapixel, for example 896x1152 (portrait, bust), 768x1344 (full body), 1536x864 (wide scenes, use cowboy shot), 1024x1024 (face)

• Text encoder: qwen_3_06b_base.safetensors; VAE: qwen_image_vae.safetensors (both from the official Anima repository, https://huggingface.co/circlestone-labs/Anima, under split_files/text_encoders/ and split_files/vae/)

Prompting

• Use Danbooru-style tags, then a plain English paragraph describing the scene, the light source, and the mood. The showcase prompts use exactly this pattern.

• No quality or score tags are needed.

• Name the light source and its color on the light itself (a lamp, a window, the sun), not on the face.

• For full-body shots, say the skirt length; use black pantyhose rather than thighhighs if you do not want stocking tops.

• For adult proportions in close shots, avoid close-up and add mature female; add big head, voluminous hair to the negative prompt if heads come out large.

• Lanterns, signs and book spines can pick up fake text. Describe them as plain or blank and add text, letters, kanji to the negative prompt.

ComfyUI

UNETLoader (put the file in models/diffusion_models) -> ModelSamplingAuraFlow (shift 3) -> KSampler; CLIPLoader with qwen_3_06b_base.safetensors, type stable_diffusion; VAELoader with qwen_image_vae.safetensors. All showcase images carry their ComfyUI graph in the PNG metadata.

Baked components (credits)

• Vetehine Dreamland ILL & ANM, v.0.1 ANM (base) by Vetehine, weight 1.0: https://civarchive.com/models/2170933?modelVersionId=3287961

• ushikani_kassen-styles, anima-base1.0 by zhihu, weight 0.6: https://civarchive.com/models/2760349?modelVersionId=3106457

• 448 - guanren79145223's Style, ANIMA by monkewt, weight 0.25: https://civarchive.com/models/2765265?modelVersionId=3112478

• CowCowCrab / Saiou-ga-ushi Style, v4.0 (Anima) by yeetyah145, weight 0.5: https://civarchive.com/models/1757934?modelVersionId=3165356

• @zxcvnmweruo Style Anima, v1.0 by Umbre99, weight 0.35: https://civarchive.com/models/2830588?modelVersionId=3193979

• [Anima]ppolar, v1.0 by ScarletDimension, weight 0.35: https://civarchive.com/models/2705088?modelVersionId=3037966

Thanks to the authors above and to CircleStone Labs for Anima.

License and use

Non-commercial. The model is a derivative of Anima (CircleStone Labs Non-Commercial License) and of the components listed above; their terms apply. Do not sell the model or offer it as a paid generation service.

中文简介

Anima 2B z3v0.1 是基于 2B Anima 结构的画风融合模型：底模 Dreamland ANM v0.1，按固定权重烤入五个 Anima 画风 LoRA，出图不必再挂这些 LoRA。推荐 30 步、CFG 4.5、er_sde + simple、AuraFlow shift 3，分辨率约 100 万像素。提示词写标签加一段英文场景描述，不需要质量词。仅限非商用。