This is an experimental LoRA series trained on the Anima-preview / original Anima 2B architecture.

Version overview

• data01-v1.0 — Pure anime dataset, based on my personally curated SDXL mixed-style dataset.

• data02-v1.0 — Another anime-focused dataset, also based on my personally curated SDXL mixed-style dataset.

• data03-v1.0 — Based on data02 with additional partial 2.5D elements, providing more volume, depth, and semi-realistic rendering when prompted appropriately.

No specific trigger word is required. Use descriptive prompts or Danbooru-style tags directly.

Model compatibility

Native:
Anima / Anima-preview / original Anima 2B architecture

Experimental:
Anima 2.9B
https://civarchive.com/models/2855007/anima-29b?modelVersionId=3224434

These LoRAs were originally trained for the Anima 2B / preview architecture. When using them with Anima 2.9B, the original 28 LoRA blocks need to be remapped to their corresponding positions in the expanded 40-block architecture.

⚠️ Anima 2.9B setup

ComfyUI

🔴 For the current ComfyUI Anima 2.9B setup, install the following patch so the expanded model can be initialized and loaded correctly:

ComfyUI-Anima-2.9B
https://github.com/gazingstars123/ComfyUI-Anima-2.9B

Install it under:

ComfyUI/custom_nodes/

and restart ComfyUI.

The patch works automatically at startup. No additional workflow node is required.

To use these original Anima 2B LoRAs with Anima 2.9B, also install:

ComfyUI-Anima-2B-LoRA-to-2.9B
https://github.com/hinablue/ComfyUI-Anima-2B-LoRA-to-2.9B

When applying these LoRAs to Anima 2.9B, use:

Anima 2B LoRA → 2.9B Loader

instead of the regular LoRA loader.

Forge Neo

Recent versions of Forge Neo support Anima directly, so the additional ComfyUI model-loader patch is not required.

To use these original Anima 2B LoRAs with Anima 2.9B in Forge Neo, install:

Anima LoRA 2B → 2.9B Remap Fix
https://github.com/re-forest/anima-lora-29b-remap-fix

Install it in the Forge Neo extensions directory and restart Forge Neo.

The extension automatically detects compatible Anima 2B LoRAs and remaps their original 28-block keys to the corresponding positions in the Anima 2.9B 40-block model at runtime.

Native or already-converted 2.9B LoRAs are detected and are not remapped.

The original LoRA and model files are not modified.

Note: Anima 2.9B compatibility is experimental. These tools provide structural compatibility by remapping the original Anima 2B LoRA blocks to their corresponding positions in the expanded 2.9B architecture. Results may therefore differ slightly from the original Anima 2B model.

Recommended settings

• LoRA Weight: 0.8–1.0
0.8–0.9 is usually the most balanced range.

• Sampler: er_sde / euler_a / Res Multistep

• Scheduler: Simple or SGM_Uniform

• Steps: 25–30

• CFG Scale: 4–5

Prompt tips

• Start with quality tags such as masterpiece, best quality, highres

• For negative prompts, score_1, score_2, score_3 is usually sufficient

• Excessive negative tags may reduce variation or over-correct details

For official Anima prompting details:

https://huggingface.co/circlestone-labs/Anima