Replies: 2 comments
-
I've had great success with textual inversion. It just might take a few tries to get the number of images, vectors and steps right. Also be sure to use a stable diffusion model that fits with the general art style, for example a cartoon/anime artist would work better with waifu diffusion. |
Beta Was this translation helpful? Give feedback.
-
For training a general style you should try DreamBooth training instead. Most of the custom models for specific artists and styles are made with with DreamBooth. DreamBooth requires a lot of vram so I would recommend using a google colab like this one if you don't have enough. If you want to keep using hypernetworks you could try changing your learning rate and step count. I get decent results with "5e-5:100, 5e-6:1500, 5e-7:10000, 5e-8:20000" to 20000 steps but you may need more if you have a large number of source images. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi everyone,
I would like to add an artist style to the model. It is a great artist we can find in 'deviantart'.
Is it possible ? If yes, how can we make that ?
I tried to train with the hypernetwork thing but it's not working. I followed some "tuto" on the net, but images are so different, I am not sure the model can learn the general "style".
I am a newbie so thanks for answers and patience.
Beta Was this translation helpful? Give feedback.
All reactions