Embedding

This tab appears only for training Embeddings.

• Base embedding (default - blank): fill it only if you want to train more to an existing embedding, leave it empty for a new embedding.
• Token Count (default - 1): The number of tokens to use for your embedding. The size of a token changes per model. 2 can be ok for simple embeddings like a face, more tokens can help to learn more like body features and will be stronger in more complex prompts (6 to 10 for example).
• Initial embedding text (default - * ): The initial embedding text used when creating a new embedding. Keep this short, either as '*' or 'a man' or 'a woman' for example to help the model understand what you are training. Remember it must use less tokens that the token count or it will get truncated.
• Embedding Weight Data Type (default - float32): What the embedding will be loaded into memory as. As embeddings are small, there is little reason to move to bfloat16, but it is an option.
• Placeholder (default - <embedding>): it must match the placeholder you are using in your images captions. If you keep the default value, your captions should use the same (photo of <embedding>) but you can use any placeholder you want like "TomCruise" with captions like "photo of TomCruise". It doesn't need to be a single word and can include spaces.