ComfyUI Embeddings Guide: Textual Inversion for Style & Quality

What are Embeddings?

Embeddings (also called Textual Inversions) are tiny model files — usually just a few KB — that encode a specific concept, style, or quality improvement into a single keyword. Think of them as compressed prompt shortcuts.

Instead of writing a long negative prompt to avoid common artifacts, you can use an embedding like EasyNegative that encodes all those avoidance rules in one word.

Where to Download Embeddings

• Civitai — Filter by type: "Textual Inversion" or "Embedding"
• HuggingFace — Search for textual inversion models

Popular embeddings for SD1.5:

Installation

1. Download the .safetensors or .pt embedding file
2. Place it in ComfyUI/models/embeddings/
3. Restart or refresh ComfyUI

ComfyUI/
├── models/
│   └── embeddings/
│       ├── EasyNegative.safetensors
│       └── bad-hands-5.pt

How to Use Embeddings

Embeddings are used directly in the CLIP Text Encode node — just type the embedding name in your prompt text:

embedding:EasyNegative

If the embedding is in a subfolder:

embedding:SD1.5/EasyNegative

Adjusting Embedding Weight

You can control how strongly the embedding affects generation using the standard weight syntax:

(embedding:EasyNegative:1.2)

Example: Using EasyNegative

Positive prompt:

1girl, portrait, detailed face, soft lighting, masterpiece

Negative prompt:

(embedding:EasyNegative:1.2), low quality, blurry

The EasyNegative embedding replaces a long list of negative quality keywords with a single token.

Autocomplete Plugin

The ComfyUI-Custom-Scripts plugin adds autocomplete support — as you type embedding: in a text field, it suggests available embedding names. This saves time and prevents typos.

Common Issues and Fixes

Embedding has no visible effect

• Verify the embedding is compatible with your model version (SD1.5 embedding with SD1.5 checkpoint)
• Check the file is in ComfyUI/models/embeddings/ and ComfyUI has been restarted
• Try increasing the weight: (embedding:EasyNegative:1.5)

"Embedding not found" or similar error

• The file name must match exactly what you type (case-sensitive on some systems)
• Refresh ComfyUI (F5) after adding new embedding files

Embedding causes artifacts

• The weight may be too high — reduce to 0.8–1.0
• The embedding may be incompatible with your specific checkpoint (some fine-tuned models conflict with certain embeddings)

Related Guides

• Prompt Guide — Write better prompts
• LoRA Guide — Style and subject fine-tuning
• Install Models — How to install different model types