This is a SD1.5 model trained on pop arts made by the one and only Roy Lichtenstein (And some other pop arts).
The model can only really do portraits, and even then, to get decent looking results, you do have to tinker with the prompt/create more samples.
Occasionally, it still makes realistic looking people with pop art background. For the time being, the best thing you can do is to simply adjust the prompt and/or rerun it.
There are definitely a lot of things I can do to improve the model, but I haven't got around to doing that.
To reference the art style, use the token: roypop style
You can check out the diffuser model here on huggingface
There is already an existing model that uses textual inversion. This is trained using Dreambooth instead, whether or not this method is better, I will let you judge.
Description
FAQ
Comments (7)
Unrelated, but Happy New Year!
Hi ItsJayQz,
I made a few Dreabooths on people and pets, and I was wondering about good settings to train a style.
How many source images you would recommend using, with how many steps to get a good result?
Thanks!
Unfortunately there isnt a straight answer for this, it really depends on the style. Normally I would have 30-50, the steps are simply the number of training images times 100.
I see so experimentation is required...
For people there is a set of regularization images, to prevent everyone from becoming the same person. Are there regularization images for style too, have you been using regularization images?
Yes there are reg images for style.
I have been using Nitrosocke's reg images, the folder name is the the class name.
https://drive.google.com/drive/folders/19pI70Ilfs0zwz1yYx-Pu8Q9vlOr9975M
Thank you very much!
@ItsJayQz GRACIAS muchas gracias
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.