Anima 2.9B z3v0.1 is a fast style merge on the 40-block Anima-2.9B architecture. It is Anima-2.9B v1.0 with the TURBO 1.1 (2.9B extract) distillation baked in at full strength, plus five Anima style LoRAs remapped from 28 to 40 blocks and baked in at fixed weights. You get the look in about 8 steps without loading any LoRA. It is the 2.9B sibling of Anima 2B z3v0.1 and carries the same five styles.

Recommended settings

• Steps: 8

• CFG: 3. At CFG 1 (the plain turbo setting) skin tends to show a glossy cell-like highlight pattern; CFG 3 with shiny skin, oily skin in the negative clears it. Negative prompts work at CFG 3.

• Sampler / scheduler: er_sde / simple

• Model sampling: AuraFlow shift 1.73

• Resolution: about 1 megapixel, e.g. 896x1152 (portrait), 1152x896 (landscape), 1024x1024

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

• Do not add a turbo LoRA, and do not load the five baked style LoRAs again.

Prompting

• No trigger words needed. Write Danbooru-style tags, then a short plain-English paragraph describing the scene, the light source and the mood. The showcase prompts use exactly this pattern.

• No quality or score tags needed.

• Name the light source on the light itself (a lamp, neon, the sun).

• For adult proportions add adult woman, mature female; add child, loli, chibi, big head to the negative. Add huge breasts to the negative if breasts come out oversized.

ComfyUI

Requires the ComfyUI-Anima-2.9B custom node (it registers the 40-block model; without it the file will not load). Then: UNETLoader (file in models/diffusion_models) -> ModelSamplingAuraFlow (shift 1.73) -> KSampler; CLIPLoader with qwen_3_06b_base.safetensors, type stable_diffusion; VAELoader with qwen_image_vae.safetensors. Ordinary 28-block Anima LoRAs need a 28->40 remap loader (e.g. AnimaLoRARemapTagLoader); native 2.9B LoRAs load directly. All showcase PNGs carry their ComfyUI graph.

Baked components (credits)

• Anima-2.9B, v1.0 (base) by gazingstars321, weight 1.0: https://civarchive.com/models/2855007?modelVersionId=3224434

• TURBO 1.1 Extracted, A1 (2.9B extract) by BlueberryTrain, weight 1.0: https://civarchive.com/models/2892819?modelVersionId=3270733

• 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

Style LoRAs were remapped 28->40 blocks (existing blocks only, new blocks left untouched). Thanks to the authors above and to CircleStone Labs for Anima.

License and use

Non-commercial. Derivative of Anima (CircleStone Labs Non-Commercial License), of Anima-2.9B, and of the components above; their terms apply. Do not sell the model or offer it as a paid generation service.

中文简介

Anima 2.9B z3v0.1 是基于 40 层 Anima-2.9B 结构的快速画风融合模型：底模 Anima-2.9B v1.0，满强度烤入 TURBO 1.1（2.9B 提取版），另按固定权重把五个 Anima 画风 LoRA 从 28 层重映射到 40 层后烤入，和 Anima 2B z3v0.1 同一套画风。需要安装 ComfyUI-Anima-2.9B 自定义节点。推荐 8 步、CFG 3、er_sde + simple、AuraFlow shift 1.73，约 100 万像素。不需要触发词，写标签加一段英文场景描述即可，不需要质量词。仅限非商用。