Embedding Tutorial

Use text embeddings for better generation control.

What is Embedding

Embedding (or Textual Inversion) is a technique to represent specific concepts, styles, or objects as custom tokens in the model's vocabulary.

Use Cases

• Add new concepts or styles
• Character consistency
• Specific art styles
• Custom objects or poses

Embedding Files

Format: .pt, .safetensors, or .png

Location: ComfyUI/models/embeddings/

Basic Usage

Step 1: Download Embedding

Get embedding files from sources like:

• Civitai
• Hugging Face
• Civitai embeddings

Step 2: Place File

ComfyUI/models/embeddings/
  embedding-name.safetensors

Step 3: Use in Prompt

# Use embedding by name
embedding:embedding-name

# Example
a person embedding:my-character

Common Embeddings

Style Embeddings

Embedding	Effect
bad-artist	Negative for bad anatomy
bad-hands	Fix hand issues
easynegative	General quality improvement
deep-negative	Stronger negatives

Character Embeddings

• Character-specific trained embeddings
• Maintain consistent appearance
• Use: embedding:character-name

Using Negative Embeddings

bad-artist-anime

# Negative prompt
(bad-artist-anime:0.8), (bad-hands:1.2)

easynegative

# Much simpler than listing all negatives
easynegative

deep-negative

# Stronger effect
deep-negative

Advanced: Custom Embeddings

Training Embedding

Requires:

• 5-20 example images
• Subject representation images
• Training script (for A1111)

Using Custom Embedding

# In prompt
my-custom-concept

# With weight
(my-custom-concept:1.2)

Tips

1. Naming: Use descriptive names
2. Organization: Group by type/style
3. Testing: Try different weights
4. Combination: Mix multiple embeddings

Troubleshooting

Issue	Solution
Not loading	Check file location
No effect	Check spelling in prompt
Too strong	Lower weight (0.5-0.8)

Next Steps

• LoRA - Use LoRA models
• ControlNet - Structure control
• Prompt Basics - Better prompts