ComfyUI-Minimax-H3-Text-Embeddings

ComfyUI MiniMax-H3 Text Embeddings

This custom node provides a single node that applies one MiniMax-H3 text embedding (safetensors) to an existing conditioning.

It is designed for the DiffSynth MiniMax-H3 text-embedding template format (Diffusion Templates), where each template file stores a weight tensor.

Node

• Name: MiniMax-H3 Apply Text Embeddings
• Input:
  • conditioning (CONDITIONING)
  • embedding_name (dropdown from models/embeddings)
  • strength (FLOAT)
• Output:
  • conditioning (CONDITIONING)

Usage

• Select one embedding file in the dropdown.
• Adjust strength as needed.
• Chain multiple nodes to apply multiple embeddings.

Original Project

• DiffSynth MiniMax-H3 Text Embeddings: https://modelscope.ai/models/DiffSynth-Studio/MiniMax-H3-Text-Embeddings

Notes

• Embedding files are loaded from ComfyUI models/embeddings.
• The loader expects a safetensors key named "weight" with shape [tokens, 5120].
• Embeddings are prepended as extra tokens to the input conditioning.
• The node also extends MiniMax token tags (minimax_token_tags) so token/tag lengths stay consistent.