The loras were trained on noobai-vpred, but also work well with noobai eps models, like yiffymixv61.

One way to use this model is to generate base image with noobai-vpred and then use lineart controlnet or simple img2img to further refine the image with eps model.

v0.02 was trained on dataset with partially removed watermark (some artifacts will still be visible in gens).

v0.01 was trained up on images with watermarks, causing them to appear in generations. As of now I have removed them, and will train a new version in due time.