ComfyUI Extension: comfyui-qwen35-anima

comfyui-qwen35-anima Introduction

The comfyui-qwen35-anima extension is a custom node for ComfyUI that enhances the capabilities of the Anima 2B diffusion model by integrating the Qwen 3.5 4B text encoder. This extension is designed to improve text understanding and processing within the Anima 2B model, which originally comes with a smaller Qwen 3 0.6B text encoder. By utilizing a hybrid architecture that combines Mamba2-style selective state space (SSM) blocks with gated self-attention, this extension offers a more robust and nuanced text encoding solution. This can be particularly beneficial for AI artists looking to generate more accurate and contextually rich outputs from their diffusion models.

How comfyui-qwen35-anima Works

At its core, the comfyui-qwen35-anima extension leverages a hybrid model architecture that alternates between selective state space (SSM) blocks and self-attention mechanisms. Imagine the SSM blocks as a way to selectively focus on important parts of the input text, much like how a spotlight highlights key areas on a stage. The self-attention layers then act like a director, coordinating these highlighted areas to ensure they work together harmoniously. This combination allows the model to better understand and encode complex text inputs, leading to more coherent and contextually appropriate outputs when used with the Anima 2B diffusion model.

comfyui-qwen35-anima Features

• Hybrid Architecture: The extension uses a combination of 24 SSM layers and 8 self-attention layers, strategically placed to maximize text encoding efficiency.
• Enhanced Text Understanding: With a hidden size of 2560 and an output dimension of 1024, the model is finely tuned to match the embedding size expected by Anima 2B, ensuring seamless integration and improved performance.
• Large Vocabulary: Supports a vocabulary of 248,320 tokens, allowing for a wide range of text inputs and nuanced language processing.
• Quantized Weights: Utilizes FP8 quantized weights with BF16 norms, optimizing the model for performance without sacrificing accuracy.

comfyui-qwen35-anima Models

The extension specifically integrates the Qwen 3.5 4B model, which is a significant upgrade from the default Qwen 3 0.6B model included with Anima 2B. This model is ideal for scenarios where enhanced text understanding is crucial, such as generating detailed and contextually rich artistic outputs. By using this model, AI artists can expect more precise and varied text-to-image translations.

What's New with comfyui-qwen35-anima

v0.3.0 — ExpRMSNorm Late Norm Fix (2026-03-09)

• Introduced a critical fix in the late normalization layer, using exp(weight) parameterization to preserve token identity, significantly improving token diversity and reducing cross-prompt similarity.

v0.2.0 — Mamba2 SSM Architecture Rewrite (2026-03-09)

• Addressed five critical bugs in the SSM block, enhancing the model's ability to process and encode text inputs accurately.

v0.1.0 — Initial Release (2026-03-09)

• Launched the custom node for the Qwen 3.5 4B hybrid text encoder, fully compatible with Anima 2B diffusion models.

Troubleshooting comfyui-qwen35-anima

If you encounter issues while using the comfyui-qwen35-anima extension, here are some common problems and solutions:

• Model Not Loading: Ensure that all required files, such as qwen35_4b.safetensors, are correctly placed in the specified directories. Double-check the file paths and restart ComfyUI.
• Unexpected Outputs: If the generated outputs are not as expected, verify that the correct model and settings are selected in the ComfyUI interface. Adjusting the text input or model parameters may also help.
• Performance Issues: Ensure your system meets the necessary requirements, such as having a compatible version of ComfyUI and sufficient computational resources.

Learn More about comfyui-qwen35-anima

For further exploration and support, consider visiting the following resources:

• cosmos-qwen3.5 on Hugging Face for more details on the Qwen 3.5 4B model.
• Anima 2B by circlestone-labs for information on the Anima 2B diffusion model.
• Community forums and tutorials on ComfyUI and diffusion models can provide additional insights and troubleshooting tips tailored to AI artists. By leveraging these resources, you can enhance your understanding and usage of the comfyui-qwen35-anima extension, unlocking new creative possibilities in your AI art projects.