DoRA is trained on the style of the artist Carles Dalmau.
DoRA weight: I recommend using the DoRA weight between 0.7 and 0.9. I personally use 0.85.
Recommended model: DoRA was trained on the AnimagineXLV31 model, so I recommend using it.
Clip skip: The clip skip of the DoRA is 2, just as specified for the AnimagineXLV31 model.
When generating an image, sometimes there may be a very noticeable noise. Noise can be fixed by upscaling image in img2img with low denoise (about 0.2 - 0.3) or just try to generate the image again on different seed. This is most likely caused by the dataset or training parameters, but I'm not sure.
Update: Noise appears because of schedule type Karras.
When writing a prompt, I recommend following the guidelines that are specified for AnimagineXLV31.
描述:
训练词语:
名称: carlesDalmauStyleLora_v10.safetensors
大小 (KB): 467829
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
