---
license: mit
---

### Wrong Embedding for Stable Diffusion 2.0

A textual inversion embedding for the `<wrong>` token, intended to be used as a negative prompt to improve the quality and fidelity of generated images.

It's recommended to use the following as a negative prompt: `in the style of <wrong>`

The training and concept of this embedding is adapted from [Nerfgun3/bad_prompt](https://huggingface.co/datasets/Nerfgun3/bad_prompt).

## Demos

For these demos, `ORIGINAL` is the base prompt, `PROMPT` corresponds to a standard negative prompt, and `TOKEN` uses `in the style of <wrong>` as the negative prompt. All three images are generated using the same latent/scheduler/seed.

![](docs/hamburger.webp)