A model trained off of James Cabello "artwork" (https://twitter.com/JamesCabello_) Does not compare to the original, but does have a unique look.
Trained on NovelAI (clone) for 10k steps off of 61 images. Compatible with other TI's and LoRAs. I recommend a weight of 1.1 or less. Works on all models, ironically worst results with the model it was trained on.
Clip skip can be 1 or 2, but who is running without it? Actually running without clip skip seems to always remove the text and occasionally get better results. Text might also get canceled out depending on other TIs and LoRAs used.
Edit: removed training data because it was first to download for some people.
Description
FAQ
Comments (5)
Wow looks amazing
If you make the first file your training data, then that's what the main download link points to. For the sake of those using download managers (or downloading into colab...) please make the first file a usable file...
Sorry, I am unable to rearrange them with the new layout on this site. I get what you mean and do apologize for the inconvenience.
The .pt file downloaded is not a .pt file but instead is a zip file with various .pt contained within, probably should re-upload the file as zip file instead of a .pt file, or upload at least one real .pt file and then an additional zip file with the additional .pt contained within.
Wow, just use 61 images, but got a really good results! Actually, I have a similar situation, I have 48 pixel art images, they are the same size (852*1386) and they are all about game scenes, could you teach me the details of your training? I have never been able to train TI or Lora well, I thought it was because I didn't have enough training data, until I saw your work, I hope I can learn tricks from you. Thank you very much!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.