A textual inversion, or embedding, teaches a model one new word that you then use in your prompt.
Typical 1
An embedding adds a single token to the prompt vocabulary, and the model paints what that token means. It does nothing until its token actually appears in your prompt. A LoRA is loaded alongside the checkpoint and affects every picture it makes; an embedding waits for its word.You can ask the horde to add the token for you and say which half of the prompt it belongs in. Put it in the negative prompt for the many embeddings that exist to describe what you do not want. Leave that unset and you type the token yourself.Strength scales how much effect the token has, and only applies when the horde is inserting it for you.A generation carries at most twenty embeddings.An embedding is named by its model page, never by a published version: unlike a LoRA, the horde has no way to be told which version of an embedding you meant.