bad_prompt_version2 is a negative embedding for text-to-image diffusion models, such as Stable Diffusion. It is trained on a dataset of images with undesirable artifacts, such as malformed hands, distorted faces, and strange compositions.
The purpose of bad_prompt_version2 is to help the model generate images that are more realistic and aesthetically pleasing.
To use bad_prompt_version2, simply add it to the negative prompt of your text-to-image diffusion model. For example, if you are using Stable Diffusion, you would add the following to your prompt:
(bad_prompt_version2:0.8)This will tell the model to avoid generating images that contain the artifacts that are present in the bad_prompt_version2 dataset.
How to use bad_prompt_version2:
If you wan to use the bad_prompt_version2 then you can follow these steps to get the best result. I have provided the examples of the bad_prompt_version2 so you can use it accordingly. Here are some examples of how to use bad_prompt_version2:
- To generate a portrait of a person with realistic hands, you could use the following bad_prompt_version2 prompt:
a portrait of a person with realistic hands (bad_prompt_version2:0.8)- To generate a landscape image with realistic clouds, you could use the following bad_prompt_version2 prompt:
a landscape image with realistic clouds (bad_prompt_version2:0.8)- To generate a product photo with a clean and professional look, you could use the following bad_prompt_version2 prompt:
a product photo of a [product name] with a clean and professional look (bad_prompt_version2:0.8)Tips to use the bad_prompt_version2:
- Start with a strength of 0.8 and adjust it as needed. A higher strength will make the model more likely to avoid the artifacts in the bad_prompt_version2 dataset, but it may also make it more difficult to generate the desired image.
What to do if I don’t getting the result I want.
- If you are not getting the results you want, try adding other negative prompts to your query. For example, if you are trying to generate an image of a person with realistic hands, you could add the following negative prompts:
(malformed hands:0.8)
(deformed hands:0.8)- You can also use bad_prompt_version2 in conjunction with positive prompts. For example, if you are trying to generate an image of a black cat, you could use the following prompt:
a black cat (bad_prompt_version2:0.8)This will tell the model to generate an image of a black cat, while avoiding the artifacts in the bad_prompt_version2 dataset.
Conclusion
bad_prompt_version2 is a powerful tool that can help you generate more realistic and aesthetically pleasing images with text-to-image diffusion models. By using bad_prompt_version2 in your prompts, you can avoid generating images with undesirable artifacts, such as malformed hands, distorted faces, and strange compositions.