Hello! This is in the form of an embedding, it is only 4 KB, I don't believe the webui will accept it if I convert to safetensors. If you know of any embeddings that are doing that could you show me? I can convert to safetensors but I would have to host this off site.

there are some uses of this in the workflows shown in the gallery below so it should be possible, and if you wanted to see the results which should show through with inpainting and other specifics, since this was encoded at the native resolution of SDXL you can use it for altering images that are already made, at the size of 1024x1024 or use the inpaint area of the same approximate size and it should have good effects.

yes I'm using it in comfyUI with XL models and it isn't throwing the errors it was for a while, saying its encoders were somehow to be used only with different size values. I don't know why they changed things I even used to be able to keep the model name as separate words 'bad' and 'x' separated with a space which I know is bad practice from a coding perspective but possibly if there is a text encoding added in some types of UI it could have the effect of including the word bad and the word x instead of a word like badx which isn't in the linguistic dictionary that would then have a more favorable result than having the word badx in it, though comfy solves this by having an embedding codeword wrapper. In automatic, it should be set with the code word in the negative prompt as "bad X" since that was the original code word but the title of the file which I changed to badX since then should also suffice in the <> brackets as per normal automatic1111 syntax. Thank you.

There might have been a reprogramming issue in automatic1111, I created this before sdxl was even integrated. If you want to try this specifically, and it won't load I would suggest trying the comfyui dualtextencoder model loader and place this as an embed on the first one and combine the second one's prompt with the first and see if this will run for you. If not it should be possible in some software since this is just encoding textual information.

This was made before Kohya had come out with a training method for SDXL embeddings, I had set the system to work on the correct size and it should function with the first text encoder, I think this was the first SDXL embedding on Civitai. It can be used on ComfyUI.

come to think of it, I made the demo images in A1111 before it even had SDXL integrated, kind of surprising it might have been the diffusers libraries were different then for text encoder processing and it just included the first TE.