Important Attribution and Credits

Anima LiquidMix is a derivative experimental model-merging project built on top of the existing Anima ecosystem.

This project does not claim authorship of the original Anima architecture, the Anima 3.8B / 52-block architecture, the Qwen3.5 conditioning approach, Expanded Conditioning, or Semantic Connector v2.

The underlying Anima model architecture and Anima 3.8B development originate from the original Anima projects:

• Anima Expanded 3.8B
  https://civitai.com/models/2910974/anima-expanded-38b

• Anima 3.8B
  https://civitai.com/models/2880480/anima-38b

• Anima 3.8B repository by lylogummy
  https://huggingface.co/lylogummy/Anima-3.8B

The current ComfyUI implementation used by this workflow, including the Anima 3.8B V2 loading path and Semantic Connector v2 inference support, is provided by:

• GumGum10 / comfyui-anima-3-8B
  https://github.com/GumGum10/comfyui-anima-3-8B

LiquidMix only adapts its experimental merged weights to work with this existing ecosystem.

Semantic Connector v2 / V2 Bundle Compatibility

Anima3.8B_LiquidMix_v4.03D.safetensors is distributed in a format compatible with the current Anima 3.8B V2 bundle loading path.

The checkpoint includes the LiquidMix Anima 3.8B / 52-block diffusion weights together with the Semantic Connector v2 bundle data required by the current V2 loader.

It can therefore be loaded directly with:

Anima38BV2Loader

from:

GumGum10/comfyui-anima-3-8B

A separate LiquidMix-specific Anima38BV2EmbeddedLoader compatibility patch is no longer required.

Current Anima 3.8B Loader

The current LiquidMix workflow uses the nodes provided directly by:

GumGum10 / comfyui-anima-3-8B

https://github.com/GumGum10/comfyui-anima-3-8B

Relevant nodes include:

Anima38BV2Loader

Anima38BV2Prompt

AnimaQwen35Loader

AnimaQwen35UnifiedPrompt

The current repository includes the Semantic Connector v2 inference implementation required by the V2 workflow.

For the current LiquidMix V2 Bundle workflow, do not install the old LiquidMix-specific Anima38BV2EmbeddedLoader.

Important Update for Older LiquidMix Instructions

Earlier LiquidMix documentation instructed users to checkout the compatibility revision:

98403d3634a9a3bf929451e0d1a362345e59d982

and then run:

install_Anima38BV2EmbeddedLoader.py

That installation procedure is now obsolete.

The current Semantic Connector v2 implementation requires files that were introduced after that older revision, including semantic_connector_v2.py.

Therefore, do not checkout the old 98403d... revision for the current workflow.

Instead, install or update to the current version of:

GumGum10/comfyui-anima-3-8B

Using the current repository is recommended rather than pinning the installation to the old LiquidMix compatibility commit.

Current Installation Procedure

New installation

cd ComfyUI/custom_nodes
git clone https://github.com/GumGum10/comfyui-anima-3-8B.git

Then install the dependencies:

python -m pip install -r ComfyUI/custom_nodes/comfyui-anima-3-8B/requirements.txt

For ComfyUI Windows Portable, use the bundled Python executable instead.

Example:

C:\ComfyUI\python_embeded\python.exe -m pip install -r C:\ComfyUI\ComfyUI\custom_nodes\comfyui-anima-3-8B\requirements.txt

Then fully restart ComfyUI.

Existing installation

If comfyui-anima-3-8B is already installed:

cd ComfyUI/custom_nodes/comfyui-anima-3-8B
git pull

Then update its requirements and restart ComfyUI.

There is no need to checkout revision 98403d..., run install_Anima38BV2EmbeddedLoader.py, or manually patch the package.

Current Semantic Connector v2 Workflow

The current Anima 3.8B LiquidMix workflow uses both:

qwen_3_06b_base.safetensors

and:

qwen35_4b.safetensors

The model path is:

Anima3.8B_LiquidMix_v4.03D.safetensors

→ Anima38BV2Loader

→ optional LoRA loader

→ KSampler

The conditioning path is:

Native Qwen 0.6B

Qwen3.5 4B

→ Anima38BV2Prompt

→ expanded

→ KSampler

Qwen3.5 4B does not simply replace the native Qwen 0.6B encoder.

Both encoders are used as part of the current Semantic Connector v2 conditioning path.