This is my first attempt at Textual Inversion in stable. I've heard through research that everyone prefers models to be trained on Pruned-1.5, so that's what this model is trained on. I'm not really sure what info to include here. Textual inversion for Tswiz.
Works well with Realistic Vision V4 as well as BadDream and Unrealistic Dream negative embeddings. Trained at .005 LR, 3000 steps, 23 images.
This is largely a test. I wanted to use someone that had a nearly infinite amount of training data to test the workflow for creating TI's. this is for experimentational and educational use ONLY. Please respect the creator and the subject.
If you appreciate the model, feel free to buy me a coffee! It would be beyond greatly appreciated :)
https://www.buymeacoffee.com/TrolikStudios
I am new to TI and posting to Civit. If there is information that you guys need to know that I didn't include, please comment and I will be sure to lyk! Specifically for the trigger word. Please please PLEASE let me know if it doesn't work.
Please use responsibly... Don't make me regret putting this out here...
Description
Base version of Taylor Swift textual inversion for SD 1.5 models.
FAQ
Comments (2)
Very good model. Congratulations! I will give
you a star rating - it is well created.
I hope you have fun creating future models for us artists to use.
-
on a side note - I am sure you are aware that certain Famous faces are already inside this A.I. program and with some simple prompts it can create their faces (and bodies) with very good details.
With that in mind, I would like to suggest to you to
test your A.I. using a prompt for creating a certain "famous Face" FIRST.
Test to see if your A.I. is able to create it good. If the images you see are not very well created, then the A.I. NEEDS further training on using a Textual-Inversion model or a Lora model or Lycoris.
As a model creator myself, I do know how much time and effort it some times takes to create a model - This tip is just so you will not waste too much of your time creating a model, which your
A.I. might be able to create very well.
Do not worry, there are TONS of other famous persons which this A.I. is still not able to create very well.
-
(Below) I will leave you with this webpage, so you can take a look at how many celebrities have their images and were trained in the Stable Diffusion program.
But even if some celebrities already have a certain number of images inside the program, it does not mean the A.I. can create them perfectly. A pre-trained model is always useful.
You also need to write a proper Prompt with certain "trigger words" which will make your A.I. create exactly what you want to see.
- https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/celebrities
Have fun with your A.I.
Cheers
MAX
Hi Max! Thanks for commenting and providing some resources to the community! I actually did know that base stable does have some good knowledge about celebrities, but I didn't realize it was that extensive! In all honesty, I just completely forgot when I was building this TI.
I'm definitely curious to test how the base 1.5 model as well as base versions of other models compare with and without the TI enabled.
In terms of prompting, you're right it is extremely important. I didn't realize how useful weights and prompting was (when creating the TI) until now. Prompting your training images correctly helps immensely!
Thanks again for the resource.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.