ComfyUI-Minimax-H3-Text-Embeddings Introduction
ComfyUI-Minimax-H3-Text-Embeddings adds a custom node called MiniMax-H3 Apply Text Embeddings. It lets you apply one MiniMax-H3 text embedding to conditioning that already exists in a ComfyUI workflow. You can use it when you want to incorporate an embedding into your prompt conditioning without replacing the rest of your workflow.
The node is designed for the DiffSynth MiniMax-H3 text-embedding template format. Each supported file is a .safetensors file containing embedding data under a key named weight.
How ComfyUI-Minimax-H3-Text-Embeddings Works
Think of conditioning as the information your workflow passes along to describe what you want to generate. This node takes that existing information, adds the selected embedding as extra tokens at the beginning, and passes the updated conditioning to the next node.
You choose an embedding file and set its strength. To use more than one embedding, connect several MiniMax-H3 Apply Text Embeddings nodes in a chain, selecting a different file in each node. Each node applies one embedding to the conditioning it receives.
The node also extends MiniMax token tags (minimax_token_tags) so the tags remain aligned with the added tokens. You do not need to manage those tag lengths yourself.
ComfyUI-Minimax-H3-Text-Embeddings Features
- Apply an embedding to existing conditioning. Connect a
CONDITIONINGoutput to the node’sconditioninginput. The node returns updatedCONDITIONINGthat you can pass to the next step in your workflow. - Choose an embedding from a dropdown. The
embedding_namecontrol lists embedding files from ComfyUI’smodels/embeddingslocation. Select the file you want this node to apply. - Adjust strength. The
strengthcontrol is a floating-point number that lets you change how strongly the selected embedding is applied. Try different values and compare your results to find a setting that suits your image. - Combine embeddings. Each node applies one file. For example, you can send the first node’s conditioning output into a second node to apply another embedding.
- Keep token tags aligned. When the embedding adds tokens, the node extends
minimax_token_tagsto match.
ComfyUI-Minimax-H3-Text-Embeddings Models
The extension does not provide a list of individual models or embedding files. Instead, its dropdown uses the files available in your ComfyUI models/embeddings location.
The supported input is a DiffSynth MiniMax-H3 text-embedding template saved as .safetensors. Its weight tensor must have the shape [tokens, 5120]. In practical terms, the file needs to contain embedding data in the format this node expects; a different safetensors file may not work simply because it has the same file extension.
For the original embedding project, see DiffSynth MiniMax-H3 Text Embeddings (https://modelscope.ai/models/DiffSynth-Studio/MiniMax-H3-Text-Embeddings).
Troubleshooting ComfyUI-Minimax-H3-Text-Embeddings
The embedding I want is not in the dropdown.
- Check that the file is in ComfyUI’s
models/embeddingslocation. - Confirm that you are looking for a
.safetensorsembedding file. - Recheck the
embedding_namedropdown for the file you intended to select.
An embedding file fails to load or apply.
- Confirm that the file is a DiffSynth MiniMax-H3 text-embedding template, rather than an unrelated safetensors file.
- Check that it contains a tensor named
weight. - Check that the tensor has the expected shape,
[tokens, 5120].
How do I apply more than one embedding?
Use one MiniMax-H3 Apply Text Embeddings node per file. Connect the conditioning output of one node to the conditioning input of the next.
Does this node create conditioning from a text prompt?
No. It takes existing CONDITIONING as input, adds the selected embedding, and outputs updated CONDITIONING.
Learn More about ComfyUI-Minimax-H3-Text-Embeddings
The DiffSynth MiniMax-H3 Text Embeddings project (https://modelscope.ai/models/DiffSynth-Studio/MiniMax-H3-Text-Embeddings) is the original project linked by the author for the text-embedding templates this node is designed to use.
